{"schema_version":"onlylabs.public_analysis_evidence.v1","title":"Arcee AI analysis evidence pack","description":"Public onlylabs evidence pack for cited agent analysis: captured pages, ranked public signals, and stored web-search provenance used by the background analysis workflow.","url":"https://onlylabs.fyi/analysis/arcee","json_url":"https://onlylabs.fyi/analysis/arcee/evidence.json","generated_at":"2026-06-28T02:18:43.672Z","org":{"slug":"arcee","name":"Arcee AI","category":"neolab","category_label":"Neolab","dossier_url":"https://onlylabs.fyi/labs/arcee"},"analysis":{"url":"https://onlylabs.fyi/analysis/arcee","json_url":"https://onlylabs.fyi/analysis/arcee/analysis.json","generated_at":"2026-06-27T19:35:38.91+00:00"},"workflow":{"version":"onlylabs-deepagents-analysis-v3","provider":"deepseek","model":"deepseek-v4-pro","agent":"deepagents","public_pack_mode":"local-pages-and-events","live_web_fetches":false,"note":"Public evidence exports do not trigger live Exa calls; stored Exa provenance is included when analysis metadata contains it."},"stats":{"pages":28,"events":140,"web":0,"evidence":88,"signal_desks":{"hiring":2,"forks":12,"releases":28,"talking":11,"repos":7},"data_radar_lanes":null,"data_radar_matches":null,"stored_analysis_evidence":93,"stored_analysis_web":5,"stored_analysis_signal_desks":{"forks":12,"repos":7,"hiring":2,"talking":11,"releases":28},"stored_analysis_data_radar_lanes":null,"stored_analysis_data_radar_matches":null},"stored_web_provenance":{"queries":["\"Arcee AI\" frontier AI lab recent model release research hiring GitHub Hugging Face","\"Arcee AI\" AI lab what they are building talking about hiring releasing forking"],"request_ids":["2d27a4c1b2c4dce99fc2740084177552","b263555ab8553a60163d0a381fb78763"],"skipped":null},"evidence":[{"ref":"P1","kind":"page","title":"arcee-ai/trinity-large-tech-report repository metadata","date":"2026-06-11T02:53:16.483891+00:00","date_source":null,"source_url":"https://github.com/arcee-ai/trinity-large-tech-report","signal_url":null,"signal_json_url":null,"text":"# arcee-ai/trinity-large-tech-report\n\nStars: 124\n\nForks: 5\n\nOpen issues: 0\n\nCreated: 2026-01-27T20:08:46Z\n\nPushed: 2026-02-19T02:01:44Z\n\nDefault branch: main\n\nFork: no\n\nArchived: no\n\nREADME:\n# Arcee Trinity Large - Technical Report\n\n<img width=\"1472\" height=\"828\" alt=\"trinity\" src=\"https://github.com/user-attachments/assets/ffca3a5f-e8cc-48a8-863d-ec2760ffaf5e\" />\n<br><br>\n\nWe present the technical report for Arcee Trinity Large, a sparse Mixture-of-Experts model with 400B total parameters and 13B activated per token. Additionally, we report on Trinity Nano and Trinity Mini, with Trinity Nano having 6B total parameters with 1B activated per token, Trinity Mini having 26B total parameters with 3B activated per token. The models’ modern architecture includes interleaved local and global attention, gated attention, depth-scaled sandwich norm, and sigmoid routing for Mixture-of-Experts. For Trinity Large, we also introduce a new MoE load balancing strategy titled Soft-clamped Momentum Expert Bias Updates (SMEBU). We train the models using the Muon optimizer. All three models completed training with zero loss spikes. Trinity Nano and Trinity Mini were pre-trained on 10 trillion tokens, and Trinity Large was pre-trained on 17 trillion tokens. The model checkpoints are available on [Hugging Face](https://huggingface.co/arcee-ai)."},{"ref":"P2","kind":"page","title":"Ai Model Routing For Maximum Savings","date":"2026-06-27T20:01:05.051724+00:00","date_source":null,"source_url":"https://www.arcee.ai/blog/ai-model-routing-for-maximum-savings","signal_url":null,"signal_json_url":null,"text":"Arcee AI | Benefits of Intelligent Model Routing: See Arcee Conductor in Action \n\nTrinity Large Thinking: Available on OpenRouter.\n\nTry now ↗\n\nENTERPRISE\n\nResearch\n\nCOMPANY\n\nGet API\n\nBlog\n/\nModel Routing for Maximum Savings\n\nModel Routing for Maximum Savings\nNora He\n,\n\nJianheng Xiao\n,\n\nAndrew Walko\n,\n\nSahana Raghuraman\n,\n\n•\n\nMarch 17, 2025\n\nFacing growing, unpredictable AI budgets? Arcee Conductor intelligently routes prompts to the optimal AI model based on complexity, cutting costs by up to 99% per prompt without sacrificing quality. Beyond a simple LLM router, it offers a comprehensive catalog of both SLMs and LLMs.\n\nAI spending represents a significant budget for all businesses these days. For many organizations, it’s also a growing and unpredictable budget, as the cost varies according to your teams’ usage of AI models. But there’s a new way to rein in this spend: instead of working only with the premium AI models like Claude or GPT-4o, now it’s easy to route your queries to the best model for that specific input. The cost savings are dramatic: the premium  AI models cost up to 188 times more than smaller models  for each prompt processed while often delivering only marginal improvements – especially for routine tasks. \nIn this article, we’ll explain how it’s now possible to ensure superior results from your AI models every time and at the lowest possible cost. \nWhat’s the best AI model for business? \nWhen choosing the best AI model, most businesses prioritize selecting one that applies to the largest number of their use cases while balancing quality and cost efficiency. However, no single model is ideal for every prompt.\nA high-powered model may offer superior output quality for complex queries, but for more straightforward routine tasks, a user pays the high cost without getting any significant value-add. Meanwhile, smaller-sized models sometimes fail to effectively handle the most complex tasks. \nBusinesses have had to choose between performance and cost-efficiency, with no real middle ground.Until now. With intelligent model routing by Arcee AI, you no longer have to work with just one model.\nWhat is Arcee Conductor? \nArcee Conductor is our intelligent "},{"ref":"P3","kind":"page","title":"Extending Afm 4 5b To 64k Context Length","date":"2026-06-27T20:01:04.257392+00:00","date_source":null,"source_url":"https://www.arcee.ai/blog/extending-afm-4-5b-to-64k-context-length","signal_url":null,"signal_json_url":null,"text":"Arcee AI | Extending AFM-4.5B to 64k Context Length \n\nTrinity Large Thinking: Available on OpenRouter.\n\nTry now ↗\n\nENTERPRISE\n\nResearch\n\nCOMPANY\n\nGet API\n\nBlog\n/\nExtending AFM-4.5B to 64k Context Length\n\nExtending AFM-4.5B to 64k Context Length\nCharles Goddard\n,\n\n•\n\nJune 23, 2025\n\nFrom 4k to 64k context through aggressive experimentation, model merging, distillation, and a concerning amount of soup.\n\nThe other day Arcee finally announced the first of our from-scratch foundation models, AFM-4.5B . Learning to train a foundation model is a long and arduous journey, and there are many lessons and learnings that we will be sharing in the full tech report in the coming weeks. In the meantime, I wanted to pull back the curtain on one particular part of the training process: extending the context length.\nWe extended AFM-4.5B from 4k to 64k context through aggressive experimentation, model merging, distillation, and a concerning amount of soup. This post will be a pretty unflattering look at the raw meat of the experimental process and the various approaches we tried, eventually arriving at a final model that performs well on both short and long context tasks. Bon appétit.\nDisclaimer : AFM-4.5B was recently introduced as a preview, with a full open-weight release (under a CC-BY-NC license) planned for early July. The evaluations presented here are based on our first checkpoint, captured immediately after the completion of our initial mid-training phase on June 3rd. The model continues to undergo additional training, including further pretraining, instruction tuning, and reinforcement learning. As a result, the benchmarks shared in this post reflect only these experiments and should be considered preliminary. Final benchmark results for the official release may differ as the model continues to improve.\nApproaches to Long Context Training \nLong-context training is a heavily researched topic, and there are many great publications that we stood on the shoulders of. Here&#x27;s a little reading list of papers we found particularly useful:\nSkyLadder: Better and Faster Pretraining via Context Window Scheduling \nd Extension of Large Languages Models \nHow to Train Long-Context L"},{"ref":"P4","kind":"page","title":"How Arcee Ai Helped Madeline Co Build A World Class Reasoning Model From First Principles","date":"2026-06-27T20:01:04.080784+00:00","date_source":null,"source_url":"https://www.arcee.ai/blog/how-arcee-ai-helped-madeline-co-build-a-world-class-reasoning-model-from-first-principles","signal_url":null,"signal_json_url":null,"text":"Arcee AI | Building Madeline-s1, a World Class Reasoning Model \n\nTrinity Large Thinking: Available on OpenRouter.\n\nTry now ↗\n\nENTERPRISE\n\nResearch\n\nCOMPANY\n\nGet API\n\nBlog\n/\nBuilding Madeline-s1, a World Class Reasoning Model\n\nBuilding Madeline-s1, a World Class Reasoning Model\nAndrew Walko\n,\n\nRaghav Ravishankar\n,\n\nPrince Rumi\n,\n\n•\n\nJune 6, 2025\n\nHow Arcee AI helped Madeline build a world-class reasoning model from first principles.\n\nMadeline & Co.’s Challenge \nMadeline & Co. is an end-to-end AI-powered strategy, design, and innovation platform that helps anyone, from in-house teams and founders to marketers and creatives, navigate complex decisions with clarity and confidence. At the core is Madeline-s1, a powerful language model trained in design, strategy, systems thinking, UX, and storytelling, delivering real-time insights and intelligent recommendations as you build.\nWhen initially building out their product suite, Madeline & Co. tried off-the-shelf large language models (LLMs); however, they constantly ran into issues of high inference costs, poor performance at scale, and inconsistent accuracy for their specific domains. The accuracy issues they faced primarily revolved around a lack of context-specific reasoning, cross-disciplinary synthesis, and brand-safe output. Madeline & Co. founder, Prince Rumi , described one specific example:\n“ When exploring a brand strategy for a sustainability startup, general models like Claude Sonnet 3.7 and GPT-4 would describe common channels or run-of-the-mill SWOTs. However, we required a model that would draw from a cross-section of startup decks, ethnographic research, brand campaigns, and founder memos to suggest an unexpected but contextually valid launch path—say, a limited-release collaboration with a fashion designer in the climate space. That leap requires nuance, not just knowledge. ”\nThis dissatisfaction with LLMs led Madeline & Co. to partner with Arcee AI in building a custom reasoning model that could reason with depth, align with internal frameworks, and operate with the flexibility needed for creative and strategic exploration. By partnering with Arcee AI, Madeline & Co. gained access to Arcee’s research l"},{"ref":"P5","kind":"page","title":"Optimizing Arcee Foundation Models On Intel Cpus","date":"2026-06-27T20:01:03.965566+00:00","date_source":null,"source_url":"https://www.arcee.ai/blog/optimizing-arcee-foundation-models-on-intel-cpus","signal_url":null,"signal_json_url":null,"text":"Arcee AI | Optimizing Arcee Foundation Models on Intel CPUs \n\nTrinity Large Thinking: Available on OpenRouter.\n\nTry now ↗\n\nENTERPRISE\n\nResearch\n\nCOMPANY\n\nGet API\n\nBlog\n/\nOptimizing Arcee Foundation Models on Intel CPUs\n\nOptimizing Arcee Foundation Models on Intel CPUs\nAndrew Walko\n,\n\nJulien Simon\n,\n\n•\n\nSeptember 30, 2025\n\nExplore how to optimize small language models on Intel’s latest CPU, utilizing Arcee AI’s AFM-4.5B and Intel-optimized inference libraries.\n\nThe size of language models is often a barrier to running and scaling AI solutions where a business needs them. The pure size of the models leads to high inference costs and large hardware requirements. Small language models address this problem by enabling individuals and companies to run AI in the most cost-effective manner possible, anywhere. As we showcased in the blog \" Is Running Language Models on CPU Really Viable? \", SLMs and CPUs make a great combination for hosting language models at the edge, offering cost-effectiveness. \nThe combination of model size, capability, and hardware optimization showcases a major step toward making advanced models deployable on affordable, widely available hardware, paving the way for new opportunities in on-device intelligence beyond the cloud.\nIn this blog, we’ll showcase how to optimize Arcee’s first foundation model, AFM-4.5B, on Intel Xeon 6 with the Intel OpenVINO toolkit and the Hugging Face Optimum Intel library.\nAFM-4.5B\nWe released AFM-4.5B in June 2025, and since then, customers have utilized it to power edge, in-environment, and agentic solutions. Outperforming all open-source models in its size range, as shown in the benchmarks below, AFM-4.5B presents itself as a top choice for edge and compute-constrained environments. \n\nAFM-4.5B benchmarks against similar sized SLMs Intel Xeon 6\nIntel Xeon 6, codenamed Granite Rapids, is Intel’s latest server processor. The Xeon 6 processor features two CPU microarchitectures: Performance Cores (P-cores), optimized for compute-intensive, vector-based workloads such as AI and HPC, and Efficiency Cores (E-cores), optimized for task-parallel, scalar-based workloads like microservices. The P-core microarchitecture, combin"},{"ref":"P6","kind":"page","title":"Introducing Arcee Supernova Medius A 14b Model That Rivals A 70b 2","date":"2026-06-27T20:01:03.848794+00:00","date_source":null,"source_url":"https://www.arcee.ai/blog/introducing-arcee-supernova-medius-a-14b-model-that-rivals-a-70b-2","signal_url":null,"signal_json_url":null,"text":"Arcee AI | Introducing SuperNova-Medius: Arcee AI&#x27;s 14B Small Language Model That Rivals a 70B \n\nTrinity Large Thinking: Available on OpenRouter.\n\nTry now ↗\n\nENTERPRISE\n\nResearch\n\nCOMPANY\n\nGet API\n\nBlog\n/\nIntroducing SuperNova-Medius: Arcee AI&#x27;s 14B Small Language Model That Rivals a 70B\n\nIntroducing SuperNova-Medius: Arcee AI&#x27;s 14B Small Language Model That Rivals a 70B\nCharles Goddard\n,\n\n•\n\nOctober 11, 2024\n\nFirst came our flagship 70B SuperNova, followed by the 8B SuperNova-Lite. Today we add to this family of superpower Small Language Models with the release of the 14B SuperNova-Medius.\n\nArcee-SuperNova-Medius is an extremely compact (14B) yet powerful language model that balances size and performance, offering capabilities closer to those of our full-size 70B SuperNova model than our 8B SuperNova-Lite variant.\nSuperNova-Medius excels at high-quality instruction-following and complex reasoning tasks, and has a deep reservoir of world knowledge. This makes it an excellent choice for a wide variety of business use cases including customer support, content creation, and advanced technical assistance.\nHow We Trained Arcee-SuperNova-Medius\nHow did we pack so much into this 14B powerhouse? The development of SuperNova-Medius was unique, to say the least. We distilled it from Llama-3.1-405B (like SuperNova and SuperNova-Lite) – except it&#x27;s from a model that uses a different architecture, so it&#x27;s no small feat.\nDetails below straight from Arcee AI&#x27;s own Charles Goddard (the founder of MergeKit ), who explains the cross-architecture distillation.\n{{tips}}\nDistillation\nThe current hotness in the world of Small Language Models (SLMs) is definitely distillation.\nDistillation is a process by which the knowledge and capabilities of a large \"teacher\" model can be transferred to a smaller \"student\" model (ideally using less compute than it would take to train the student model from scratch). There are a number of approaches to this, which I&#x27;ll briefly cover here.\nSynthetic Data Distillation\nOften when people refer to distillation, they are referring to the process of generating synthetic data from a large model and using it to train a smal"},{"ref":"P7","kind":"page","title":"Introducing Arcee Vylinh A Powerful 3b Parameter Vietnamese Language Model","date":"2026-06-27T20:01:03.736908+00:00","date_source":null,"source_url":"https://www.arcee.ai/blog/introducing-arcee-vylinh-a-powerful-3b-parameter-vietnamese-language-model","signal_url":null,"signal_json_url":null,"text":"Arcee AI | Introducing Arcee-VyLinh - A Powerful 3B Parameter Vietnamese Language SLM \n\nTrinity Large Thinking: Available on OpenRouter.\n\nTry now ↗\n\nENTERPRISE\n\nResearch\n\nCOMPANY\n\nGet API\n\nBlog\n/\nIntroducing Arcee-VyLinh - A Powerful 3B Parameter Vietnamese Language SLM\n\nIntroducing Arcee-VyLinh - A Powerful 3B Parameter Vietnamese Language SLM\nQuan Nguyen\n,\n\n•\n\nNovember 7, 2024\n\nToday Arcee AI makes our latest contribution to AI in underserved global languages with the release of a 3B Vietnamese SLM, Arcee-VyLinh.\n\nXin Chao Vietnam! (Hello, Vietnam!)\nIn the rapidly evolving landscape of AI, Vietnamese language processing has lagged behind its global counterparts. While large language models (LLMs) have transformed how we interact with AI in English and several other languages, Vietnamese speakers have had to settle for either generic multilingual models or specialized models that compromise on performance.\nToday, we&#x27;re changing that narrative with Arcee-VyLinh, a breakthrough 3B parameter small language model (SLM) that redefines what&#x27;s possible with Vietnamese language AI. Despite its remarkably compact size, VyLinh demonstrates capabilities that surpass significantly larger models, marking a new chapter in Vietnamese NLP.\nOur journey began with a simple yet ambitious goal: to create a Vietnamese language model that could deliver state-of-the-art performance without requiring massive computational resources. We believed that with the right training approach, we could push the boundaries of what&#x27;s possible with a 3B parameter model.\nUse Cases\nBefore we dive into how we trained Arcee-VyLinh, we&#x27;d like to highlight some of the many practical use cases of the model. The advanced Vietnamese language capabilities make it versatile for both enterprise and personal applications.\nIn business settings, it excels at customer service automation, content creation, and document processing. Educational institutions can leverage it for language learning and academic writing support.\nFor developers and researchers, it serves as a powerful tool for building Vietnamese-language applications, from chatbots to content moderation systems.\nFinally, the quantized "},{"ref":"P8","kind":"page","title":"Introducing Arcees Slm Adaptation System","date":"2026-06-27T20:01:03.713947+00:00","date_source":null,"source_url":"https://www.arcee.ai/blog/introducing-arcees-slm-adaptation-system","signal_url":null,"signal_json_url":null,"text":"Arcee AI | Introducing Arcee’s SLM Adaptation System \n\nTrinity Large Thinking: Available on OpenRouter.\n\nTry now ↗\n\nENTERPRISE\n\nResearch\n\nCOMPANY\n\nGet API\n\nBlog\n/\nIntroducing Arcee’s SLM Adaptation System\n\nIntroducing Arcee’s SLM Adaptation System\nMark McQuade\n,\n\n•\n\nFebruary 7, 2024\n\nAt Arcee, we believe in a world of smaller, specialized models that we call SLM’s. The “S” stands for smaller, specialized, scalable, and secure. These models are grounded on your data, run entirely in your own environment, and are infinitely scalable for all your use cases.\n\nAt Arcee, we believe in a world of smaller, specialized models that we call SLM’s. The “S” stands for smaller, specialized, scalable, and secure. These models are grounded on your data, run entirely in your own environment, and are infinitely scalable for all your use cases.\nWe feel these SLM’s are better for 99% of business use cases. This strong belief is coupled with the confidence that training a model on your private data requires multiple layers of depth to achieve the results and outcomes you need for your model. Recent findings have highlighted that mere Simple Instruction tuning or LORA-type fine-tuning may not suffice, particularly in specific and knowledge-rich domains. To enhance our understanding and effectiveness in these areas, it&#x27;s becoming increasingly clear that we need to advance beyond these methods and come up with innovative approaches to enhance the knowledge probing.\nOur approach to domain adaptation involves a structured, four-layer process. At each of these layers, we focus on progressively enhancing the domain knowledge of the model. This systematic method ensures that the model continually advances in intelligence, specifically in relation to your private data, achieving a deeper and more refined understanding at every step.\nIn this article, we will walk through each layer of what we call our SLM Adaptation System.\n\nLayer 1: Domain Adaptive Continual Pretraining\nIn our initial phase at Arcee, we select an open-source foundational model, such as Mistral or Llama, to conduct domain adaptive continual pretraining. This process involves self-supervised training, focusing on the trad"},{"ref":"P9","kind":"page","title":"Announcing The Arcee Foundation Model Family","date":"2026-06-27T20:01:03.459126+00:00","date_source":null,"source_url":"https://www.arcee.ai/blog/announcing-the-arcee-foundation-model-family","signal_url":null,"signal_json_url":null,"text":"Arcee AI | Announcing Arcee Foundation Models \n\nTrinity Large Thinking: Available on OpenRouter.\n\nTry now ↗\n\nENTERPRISE\n\nResearch\n\nCOMPANY\n\nGet API\n\nBlog\n/\nAnnouncing Arcee Foundation Models\n\nAnnouncing Arcee Foundation Models\nMark McQuade\n,\n\nLucas Atkins\n,\n\nFernando Fernandes Neto\n,\n\nCharles Goddard\n,\n\nVarun Singh\n,\n\nJulien Simon\n,\n\n•\n\nJune 18, 2025\n\nThe first release—AFM-4.5B—is a 4.5-billion-parameter model that delivers excellent accuracy, strict compliance, and very high cost-efficiency.\n\nToday, we’re thrilled to unveil the Arcee Foundation Models , a new family of generative AI models built from the ground up for enterprise reality. The first release— AFM-4.5B —is a 4.5-billion-parameter frontier model that delivers excellent accuracy, strict compliance, and very high cost-efficiency. In short: enterprise-grade intelligence that can run anywhere—on a smartphone, at the edge, or in the cloud.\nFor a quick taste, you can test AFM-4.5B in our playground and on Together.ai .\nFor a deeper dive into the model’s training pipeline and benchmarks, details are available in our technical blog post .\nWhy did we build AFM?\nIn short, because our customers have told us they need it. Over the last 12 months we have met with more than 150 companies - from the Fortune 100 to AI startups like ourselves - to understand their challenges and evaluate how AI and small language models (SLMs) can solve them. Across hundreds of conversations and active collaborations with our customers, we repeatedly heard similar common roadblocks in their path to adopt generative AI.\nMany of our enterprise customers adopted large language models (LLMs) from providers such as OpenAI, Anthropic, and DeepSeek due to their ease of use, speed of deployment, and broad general capabilities. However, these models present significant challenges: they are expensive to operate at scale and difficult, or prohibitively costly, to customize for use cases that require in-depth domain knowledge. They also raise substantial concerns around data privacy, IP liability, and regulatory compliance, as most, if not all, LLMs are tainted with copyrighted or paywalled data sources, which may expose the business to legal o"},{"ref":"P10","kind":"page","title":"Announcing Distillkit","date":"2026-06-27T20:01:03.452286+00:00","date_source":null,"source_url":"https://www.arcee.ai/blog/announcing-distillkit","signal_url":null,"signal_json_url":null,"text":"Arcee AI | Announcing DistillKit for creating & distributing SLMs \n\nTrinity Large Thinking: Available on OpenRouter.\n\nTry now ↗\n\nENTERPRISE\n\nResearch\n\nCOMPANY\n\nGet API\n\nBlog\n/\nDistilling LLMs with Compact, Powerful Models for Everyone: Introducing DistillKit by Arcee AI\n\nDistilling LLMs with Compact, Powerful Models for Everyone: Introducing DistillKit by Arcee AI\nLucas Atkins\n,\n\nFernando Fernandes\n,\n\n•\n\nAugust 1, 2024\n\nFirst, Arcee AI revolutionized Small Language Models (SLMs) with Model Merging and the open-source repo MergeKit. Today we bring you another leap forward in the creation and distribution of SLMs with an open soure tool we&#x27;re calling DistillKit.\n\nAt Arcee AI, we&#x27;re on a mission to make artificial intelligence more accessible and efficient. Today, we&#x27;re thrilled to announce the release of DistillKit , our new open-source tool that&#x27;s set to change how we at Arcee AI create and distribute Small Language Models (SLMs).\nWhat is DistillKit?\nDistillKit is our open-source project focused on something called \"model distillation.\"\nThink of it like this: we have a big, smart model (let&#x27;s call it the teacher) that knows a lot but requires a lot of resources to run. What we want is a smaller model (the student) that can learn most of what the big model knows, but can run on your laptop or phone.\nThat&#x27;s what DistillKit does – it helps create smaller models that are powerful like the big ones, but need much less computing power. This means more people can use advanced models in more places.\n{{tips}}\nHow DistillKit Works: Teaching a Smaller Model\nWe&#x27;re using two main methods in DistillKit to transfer knowledge from the big AI to the smaller one:\nLogit-based Distillation \nThis method is like having the big model show its work to the smaller model. The smaller AI doesn&#x27;t just learn the right answers – it also learns how confident the big model is about different possible answers. This helps the smaller model think more like the bigger one.\nHidden States-based Distillation \nThis approach is about teaching the smaller model to understand information in a similar way to the big model. It&#x27;s like teaching someone to fish inst"},{"ref":"P11","kind":"page","title":"Llama 3 1 Arcee Training Support","date":"2026-06-27T20:01:03.402361+00:00","date_source":null,"source_url":"https://www.arcee.ai/blog/llama-3-1-arcee-training-support","signal_url":null,"signal_json_url":null,"text":"Arcee AI | Train, Merge, & Domain-Adapt Llama-3.1 with Arcee AI \n\nTrinity Large Thinking: Available on OpenRouter.\n\nTry now ↗\n\nENTERPRISE\n\nResearch\n\nCOMPANY\n\nGet API\n\nBlog\n/\nTrain, Merge, & Domain-Adapt Llama-3.1 with Arcee AI\n\nTrain, Merge, & Domain-Adapt Llama-3.1 with Arcee AI\nJacob Solawetz\n,\n\nJulien Simon\n,\n\n•\n\nJuly 24, 2024\n\nGet Llama-3.1 but better – customize the OS model for all your needs, using Arcee AI&#x27;s training, merging, and adaptation techniques and tools. Our team created this guide to get you started.\n\nWe are excited to announce Llama-3.1 training and merging support within Arcee Cloud , Arcee Enterprise (VPC) , and <span class=\"encased\"> mergekit </span>.\nLlama-3.1&#x27;s 128K context practically solves many problems for our customers who build domain-adapted Small Language Models (SLMs) but still need the longer context they&#x27;re accustomed to using with general model APIs.\nThe community had previously worked on longer-context SLM models, but Llama-3.1 solidifies a good pre-trained standard for us all to work from.\nThe best part about Llama-3.1 is that you can keep training it!\nAnd as we always recommend, you should undoubtedly merge it.\nContinuously Pre-Train Llama-3.1 on Arcee Cloud+VPC\nContinuous Pre-Training involves retraining a language model&#x27;s next token prediction on your proprietary set of text. This allows you to extend Llama&#x27;s knowledge of tokens within its parameters, not externally , with prompting, reasoning, or Retrieval-Augmented Generation (RAG).\nFor Llama-3.1, you can Continuously Pre-Train with a smaller context window (e.g., 8K token stacks) and merge back into the extended context window without knowledge transfer loss.\nTo reduce training time without compromising quality, we scanned LLama-3.1 with Arcee Spectrum and integrated the model into the Arcee Continuous Pre-Training routines.\nFirst, I inject a few text tokens from our internal Arcee Slack.\n\nArcee Slack Tokens Then, I run Continuous Pre-Training on Llama-3.1-8B.\n\nContinual Pre-Training status We can observe knowledge injection of Arcee Slack tokens, even with a small sample set of tokens.\n\nLlama3.1 CPT See here for more Arcee Pre-Training docs .\n"},{"ref":"P12","kind":"page","title":"Introducing The Trinity Builders Program","date":"2026-06-27T20:01:03.349049+00:00","date_source":null,"source_url":"https://www.arcee.ai/blog/introducing-the-trinity-builders-program","signal_url":null,"signal_json_url":null,"text":"Arcee AI | Introducing the Trinity Builders Program \n\nTrinity Large Thinking: Available on OpenRouter.\n\nTry now ↗\n\nENTERPRISE\n\nResearch\n\nCOMPANY\n\nGet API\n\nBlog\n/\nIntroducing the Trinity Builders Program\n\nIntroducing the Trinity Builders Program\nAnneketh Vij\n,\n\n•\n\nApril 14, 2026\n\nA community credit grant for developers, researchers, and open source builders working with Trinity models. Apply for free inference access on the Arcee API.\n\nSince we released Trinity-Large-Thinking, developers have taken the model further than we expected. Agent loops running dozens of turns, research prototypes pushing sparse MoE reasoning, open source projects built entirely around Trinity&#x27;s tool-calling capabilities.\n​That momentum has also shown up in our support queue. We increasingly hear from builders who have clear projects, clear technical plans, and real community impact but limited inference budget.\n​Today we&#x27;re launching the Trinity Builders Program: A community credit grant that gives developers, researchers, and open source builders free inference access to Trinity models on our API.\nWhy we&#x27;re doing this\n​We built Trinity so developers could own their models. Apache 2.0 weights, open architecture, full control. But open weights alone are not enough. Teams still need compute to run experiments, iterate on evals, and ship real systems.\n​We want to lower that barrier. If you&#x27;re building something meaningful with Trinity - A research project, an open source tool, a prototype that could become a product. We want to give you the compute to see it through.\n​We also have a selfish reason. The builders who run our models in real systems are the ones who surface the insights that shape what comes next. Community feedback on Preview directly informed the direction of Trinity-Large-Thinking, our reasoning-optimized variant post-trained with extended chain-of-thought and agentic RL, purpose-built for the multi-step workflows developers were already running. That feedback loop matters to us, and we want to keep it running.\n​The program covers the Trinity family models. \nWhat you get\n​Accepted applicants receive free API credits for Trinity models, allocated based on"},{"ref":"P13","kind":"page","title":"Research Spotlight 3 Learnings From 3 Use Cases Of Mergekit","date":"2026-06-27T20:01:03.281434+00:00","date_source":null,"source_url":"https://www.arcee.ai/blog/research-spotlight-3-learnings-from-3-use-cases-of-mergekit","signal_url":null,"signal_json_url":null,"text":"Arcee AI | Research Spotlight: 3 Learnings from 3 MergeKit Use Cases \n\nTrinity Large Thinking: Available on OpenRouter.\n\nTry now ↗\n\nENTERPRISE\n\nResearch\n\nCOMPANY\n\nGet API\n\nBlog\n/\nResearch Spotlight: 3 Learnings from 3 MergeKit Use Cases\n\nResearch Spotlight: 3 Learnings from 3 MergeKit Use Cases\nMariam Jabara\n,\n\n•\n\nJune 24, 2025\n\nMerging for pre-training, data privacy in healthcare, and language support\n\nMergeKit is the industry-leading tool for Model Merging, a technique that enables you to combine several pre-trained models into a smaller, more efficient model without requiring additional training (and without the need for a GPU). Merging preserves the original capabilities of models while enhancing AI performance and versatility. Arcee designed MergeKit to be both powerful and accessible, putting the power of model merging into the hands of builders of all levels.\nIn this article, we’ll distill the lessons learned from three use cases of model merging published in research papers. Whether you’re interested in improving accuracy, creating domain-specific models, or incorporating advanced reasoning capabilities into diverse language models, we’ll share how model merging plays an important role in these use cases. \nModel Merging in Pre-training of Large Language Models \nWhile model merging is often discussed in the context of post-training , this paper&#x27;s authors investigate the use of model merging techniques during the pre-training process, a largely unexplored area of research. Pre-training merging typically involves merging checkpoints from a single training run. However, researchers’ access to these intermediate checkpoints is limited, even with open models, which limits their understanding of the utility of merging in pre-training. Builders of DeepSeek and LLaMA-3 leveraged model merging in the development of their models, but they didn’t disclose the exact techniques. \nIn this work, the authors introduce the concept of Pre-trained Model Average (PMA), a strategy for merging model-level weights during pre-training. Across a diverse set of LLMs of varying sizes and architectures, the authors found that merging checkpoints from the stable training phase r"},{"ref":"P14","kind":"page","title":"Releasing Five New Open Weights Models","date":"2026-06-27T20:01:03.082675+00:00","date_source":null,"source_url":"https://www.arcee.ai/blog/releasing-five-new-open-weights-models","signal_url":null,"signal_json_url":null,"text":"Arcee AI | Arcee AI Releases Five New Open Weights Models \n\nTrinity Large Thinking: Available on OpenRouter.\n\nTry now ↗\n\nENTERPRISE\n\nResearch\n\nCOMPANY\n\nGet API\n\nBlog\n/\nReleasing Five Open-Weights Models\n\nReleasing Five Open-Weights Models\nJulien Simon\n,\n\n•\n\nJune 30, 2025\n\nSuperNova 70B, Virtuoso-Large 72B, Caller 32B, GLM-4-32B-Base-32K, and Homunculus 12B\n\nToday, we&#x27;re happy to announce the open-weights release of five language models, including three enterprise-grade production models that have been powering customer workloads through our SaaS platform and two cutting-edge research models. This release underscores Arcee AI&#x27;s fundamental commitment to democratizing access to cutting-edge AI technology through open-weight models, even for our most advanced commercial offerings.\nAt Arcee AI, we believe that the future of AI lies in transparency, accessibility, and community-driven innovation. By releasing these production-tested models as open weights, we&#x27;re enabling developers, researchers, and enterprises to deploy, customize, and build upon our work without restrictions, whether for research, commercial applications, or further model development.\nThis significant release marks our transition to the primary focus on the Arcee Foundation Model (AFM) family , which represents our next-generation approach to building efficient, compliant, and deployable foundation models. With the release of the first AFM model, AFM-4.5B-Preview , we&#x27;re confident in opening our previous generation of specialized models to the broader community.\nProduction Models \nOur production models have been battle-tested in real-world enterprise environments, delivering reliable performance across diverse use cases from customer service automation to complex reasoning tasks. These models were previously available exclusively through Arcee Conductor, our SaaS platform , and have now been released with full commercial licensing, enabling unrestricted deployment.\nArcee-SuperNova-v1\nArcee-SuperNova-v1 (70B) is a merged model built from multiple advanced training approaches. At its core is a distilled version of Llama-3.1-405B-Instruct, converted into Llama-3.1-70B-Instruct usin"},{"ref":"P15","kind":"page","title":"Meet Mergekit V0 1 Arcee Fusion Expanded Model Support Multi Gpu Acceleration","date":"2026-06-27T20:01:02.999125+00:00","date_source":null,"source_url":"https://www.arcee.ai/blog/meet-mergekit-v0-1-arcee-fusion-expanded-model-support-multi-gpu-acceleration","signal_url":null,"signal_json_url":null,"text":"Arcee AI | Meet MergeKit v0.1: Expanded Model Support, Arcee Fusion, & Multi-GPU Acceleration \n\nTrinity Large Thinking: Available on OpenRouter.\n\nTry now ↗\n\nENTERPRISE\n\nResearch\n\nCOMPANY\n\nGet API\n\nBlog\n/\nMeet MergeKit v0.1: Expanded Model Support, Arcee Fusion, & Multi-GPU Acceleration\n\nMeet MergeKit v0.1: Expanded Model Support, Arcee Fusion, & Multi-GPU Acceleration\nCharles Goddard\n,\n\nLucas Atkins\n,\n\n•\n\nFebruary 5, 2025\n\nMergeKit changed the game when it came to model merging, and today we&#x27;re excited to bring you some game-changing updates to MergeKit–with what we&#x27;re calling MergeKit v0.1. Starting today, you&#x27;ll be able to unlock the power of model merging more than ever, with enterprise hosting, premium features, and expert support.\n\nIt&#x27;s been just slightly over a year since Arcee AI acquired MergeKit and joined forces with its creator, Charles Goddard. Since then, we&#x27;ve had an incredible year of constant innovation, collaboration with the open-source community, and productizing of model merging as we built out our world-class model training pipeline. \nTo mark the one-year anniversary of Arcee AI + MergeKit, we&#x27;re bringing you the most significant updates to MergeKit to date. Check them out and let us know what you think and what you build. And as always, happy merging! \nExpanded Model Support: Merge Anything, Merge Faster\nMeet MergeKit v0.1, which dramatically expands the range of models you can merge. No longer are you limited to specific architectures explicitly supported by MergeKit. This release introduces two game-changing improvements:\nArbitrary transformers models: MergeKit now seamlessly handles any model architecture supported by the popular transformers library. This means you can merge cutting-edge models as soon as they&#x27;re released, including vision-language models like LLaVa or QwenVL, alongside the diverse collection of decoder-only models already supported. No more waiting for MergeKit to \"catch up\"–you&#x27;re empowered to merge immediately.\nRaw PyTorch Models: Beyond transformers models, MergeKit now supports merging raw PyTorch models. This opens up a world of possibilities, allowing you to merge models se"},{"ref":"P16","kind":"page","title":"Mergekit Returns To Its Roots","date":"2026-06-27T20:01:02.996951+00:00","date_source":null,"source_url":"https://www.arcee.ai/blog/mergekit-returns-to-its-roots","signal_url":null,"signal_json_url":null,"text":"Arcee AI | Mergekit Returns To Its Roots \n\nTrinity Large Thinking: Available on OpenRouter.\n\nTry now ↗\n\nENTERPRISE\n\nResearch\n\nCOMPANY\n\nGet API\n\nBlog\n/\nMergekit Returns To Its Roots\n\nMergekit Returns To Its Roots\nLucas Atkins\n,\n\n•\n\nOctober 31, 2025\n\nEffective Friday, October 31, 2025, we are returning Mergekit to the GNU Lesser General Public License v3.\n\nWhy we tried BSL\nMergekit began as a research tool for model merging and post-training. Over time, advanced techniques like fusion and tokenizer surgery lived in a private Arcee fork. We wanted to share that work and invest more in the public library, while preventing large competitors from hosting or commercializing it without engagement.\nThe Business Source License looked like a clean middle path: open enough for the community, protected enough for us. In theory, most developers could continue as before. In practice, any custom license creates gray areas. Even when the answer is \"you&#x27;re fine,\" the question itself slows adoption. Engineering teams need to route license reviews through legal. Startups hesitate. Contributors wonder if their patches will complicate future use.\nThat uncertainty did not align with how we want to build.\nWhat changed\nDuring the same period, our scope expanded beyond post-training into the full model lifecycle. We released AFM-4.5B in late July under Apache-2.0. The next generation of models ships in the same permissive terms within weeks. Mergekit now sits inside a much larger training and deployment pipeline. Keeping the library under a custom license while our models use a standard one created misalignment.\nCommunity feedback made the decision straightforward. Developers told us directly: the BSL introduced friction they did not want to navigate. They valued clarity over cleverness. We listened.\nWhat LGPL v3 means in practice\nYou can use Mergekit freely in commercial or proprietary products. If you modify and distribute the library itself, you must release those changes under LGPL v3. Keep copyright notices intact and provide library source when distributing binaries.\nFor most teams, this changes nothing operationally. For everyone, it removes the license question from the adop"},{"ref":"P17","kind":"page","title":"March Is Merge Madness","date":"2026-06-27T20:01:02.889184+00:00","date_source":null,"source_url":"https://www.arcee.ai/blog/march-is-merge-madness","signal_url":null,"signal_json_url":null,"text":"Arcee AI | March is Merge Madness \n\nTrinity Large Thinking: Available on OpenRouter.\n\nTry now ↗\n\nENTERPRISE\n\nResearch\n\nCOMPANY\n\nGet API\n\nBlog\n/\nMarch is Merge Madness\n\nMarch is Merge Madness\nMary MacCarthy\n,\n\n•\n\nMarch 4, 2024\n\nTo celebrate Arcee’s recent merger with mergekit, we’re bringing you a month of resources and knowledge on model merging.\n\nIt will be everything you&#x27;ve wanted to know about model merging, and more. Over the next four weeks, we’ll be bringing you tips and tutorials on:\nHow model merging works\nThe massive $$$ savings of model merging\nIndustry verticals where model merging is particularly useful \nWhy Arcee and mergekit joined forces. \n\nHave questions about model merging? Send them to our team on X or LinkedIn . And check out this interview with our CEO Mark McQuade who explains why Arcee is committed to becoming the leader in model merging, and how model merging fits into Arcee’s Small Language Model (SLM) system.\nTRANSCRIPT : \nMark McQuade, Arcee CEO \nWe created a larger system called an SLM adaptation system, and it has multiple layers of domain adaptation, one being Continual Pre-Training, then Alignment, which is Supervised Fine-Tuning, and then finally Retrieval Augmented Generation.\nIt&#x27;s kind of those three pillars that are our SLM Adaptation System.\nSo where model merging and mergekit fit into our system is in the very first layer in Continual Pre-training – while you have models that you can train,smaller models that you can train on a much more efficient pace and then you can merge them with larger models, right? So in the world of Continual Pre-Training that exists today, you have to train over the entire model, right? Now… we&#x27;re saying – don&#x27;t trainover the entire  model, train a much much smaller model and then merge it with a much larger model or, you know, SLM kind of stands for small… So we don&#x27;t actually believe in that large of a model… We think that 99% of business use cases can be solved with smaller models.\nSo it&#x27;s just a matter of training that smaller model and merging it with a larger model to allow for the flexibility to have, you know, a great performing model that you did not need to tra"},{"ref":"P18","kind":"page","title":"Should You Use Enterprise Llms For Your Organization","date":"2026-06-27T20:01:02.632522+00:00","date_source":null,"source_url":"https://www.arcee.ai/blog/should-you-use-enterprise-llms-for-your-organization","signal_url":null,"signal_json_url":null,"text":"Arcee AI | Should You Use Enterprise LLMs for Your Organization? \n\nTrinity Large Thinking: Available on OpenRouter.\n\nTry now ↗\n\nENTERPRISE\n\nResearch\n\nCOMPANY\n\nGet API\n\nBlog\n/\nShould You Use Enterprise LLMs for Your Organization?\n\nShould You Use Enterprise LLMs for Your Organization?\nNora He\n,\n\n•\n\nJanuary 13, 2025\n\nThe days of general-use large language models (LLMs) in companies could be coming to an end. The new reality: the use of enterprise LLMs and/or small language models (SLMs) is on the rise, as businesses realize that that impactful, efficient AI must be based on tailored models that have been customized to their own data and use cases.\n\nArtificial intelligence (AI) is transforming industries—but how can businesses leverage large language models for real results?\nEverywhere you look, AI technologies are reshaping business operations. At the core of this transformation are large language models (LLMs), which can analyze huge amounts of data, understand natural language, and generate human-like responses to user queries.\nBut as promising as LLMs are, enterprises face challenges like handling sensitive data, integrating with complex systems, and ensuring that solutions provide relevant and up-to-date information tailored to specific industries. Luckily, enterprise LLMs offer the customization and scalability businesses need to turn cutting-edge AI into practical solutions.\nIn this article, we’ll explore what makes enterprise LLMs unique, their real-world applications, and how transitioning to Small Language Models (SLMs) can bring additional value to your AI strategy.\nWhat is an Enterprise LLM?\nAn enterprise LLM is a large language model tailored to meet the needs of enterprise systems. Unlike off-the-shelf LLMs, these models can:\nHandle industry-specific requirements\nIntegrate with existing enterprise systems\nAddress challenges like data security and sensitive data management\n\nBy customizing their training data and workflows, enterprise LLMs can provide more relevant information, adapt to unique business contexts, and support decision-making at scale. These models empower organizations to use generative AI for practical, impactful applications.\nFor example"},{"ref":"P19","kind":"page","title":"Small Language Models Rising As Arcee Ai Lands 24m Series A","date":"2026-06-27T20:01:02.623393+00:00","date_source":null,"source_url":"https://www.arcee.ai/blog/small-language-models-rising-as-arcee-ai-lands-24m-series-a","signal_url":null,"signal_json_url":null,"text":"Arcee AI | Small language models rising as Arcee AI lands $24M Series A \n\nTrinity Large Thinking: Available on OpenRouter.\n\nTry now ↗\n\nENTERPRISE\n\nResearch\n\nCOMPANY\n\nGet API\n\nBlog\n/\nSmall Language Models Rising as Arcee AI lands $24M Series A\n\nSmall Language Models Rising as Arcee AI lands $24M Series A\nVentureBeat\n,\n\n•\n\nJuly 16, 2024\n\nArcee AI is enabling small language models with a $24M Series A funding round and the launch of Arcee Cloud. This innovative platform offers a hosted SaaS version of their AI, complementing the in-VPC Arcee Enterprise.\n\nThe trend toward small language models is accelerating as Arcee AI announced its $24M Series A funding only 6 months after announcing its $5.5M seed round in January 2024. The company also announced the launch of Arcee Cloud, a hosted SaaS version of their platform. This new offering complements their existing in-VPC deployment option, Arcee Enterprise. \nThe new round, led by Emergence Capital, signals growing investor confidence in the potential of smaller, more efficient AI models. \"The Series A gives us the resources to bring our solution to the masses via our new cloud platform,\" said Arcee AI Co-Founder and CEO Mark McQuade in an exclusive interview with VentureBeat. \nSmall language models (SLMs) are quickly becoming a go-to solution for enterprises in specific domains, particularly for question-answering applications.  “If you want a model for your HR use case, you don&#x27;t care that it knows who won the Academy Awards for Best Picture in 1967,\" McQuade said. \"We&#x27;ve seen great success with models that are as small as 7 billion parameters.\"\nMcQuade highlighted several use cases, including tax assistance, educational support, HR inquiries, and medical question-answering. Unlike data extraction or automated analysis tasks, these applications focus on providing accurate, context-aware responses to user queries. The versatility of SLMs in handling these specialized Q&A tasks efficiently makes them attractive across diverse industries, from finance to healthcare.\nThe rapid rise of SLMs \nAs we noted back in April , SLMs are beginning to challenge the \"bigger is always better\" approach, offering benefits in co"},{"ref":"P20","kind":"page","title":"The Hidden Obstacles Of Domain Adaptation In Llms","date":"2026-06-27T20:01:02.516249+00:00","date_source":null,"source_url":"https://www.arcee.ai/blog/the-hidden-obstacles-of-domain-adaptation-in-llms","signal_url":null,"signal_json_url":null,"text":"Arcee AI | The Hidden Challenges of Domain-Adapting LLMs \n\nTrinity Large Thinking: Available on OpenRouter.\n\nTry now ↗\n\nENTERPRISE\n\nResearch\n\nCOMPANY\n\nGet API\n\nBlog\n/\nThe Hidden Challenges of Domain-Adapting LLMs\n\nThe Hidden Challenges of Domain-Adapting LLMs\nMalikeh Ehghaghi\n,\n\n•\n\nJune 25, 2024\n\nAdapting an LLM to a specific domain might sound straightforward, but it in fact opens a Pandora&#x27;s box of challenges. Our research team explains the shortfalls of some of the most common techniques.\n\nAdapting a Large Language Model (LLM) to a specific domain seems like a fairly straightforward task: just get the model to learn some additional data, right? \nThe reality, of course, is much more nuanced than that. \nHere at Arcee AI, we&#x27;re the leaders in building and deploying highly customized language models —what we refer to as Small Language Models or SLMs (even though they’re generally not actually that small 😂).\nWe’ve designed a domain adaptation pipeline that solves for the many challenges that tend to arise when you further train an LLM to customize it for a specific topic or sector. \nAlong the way, we’ve learned a lot about what can go wrong in domain adaptation attempts – and we’re sharing those with you here.\nNavigating the Challenges of Fine-Tuning Pre-Trained Foundation Language Models\nFine-tuning of the pre-trained foundation language models has been de facto the standard approach for many years. But even in the presence of training data for fine-tuning, this process poses many challenges – and often, straightforward implementation results in poor model performance.\n\"Oops, I Forgot I Knew That!\" – The Dilemma of Catastrophic Forgetting\nFirst of all, the modern generation of LLMs already incorporates a lot of generic reasoning and world knowledge skills. \nClassical fine-tuning, during which all model weights are updated to learn from domain-specific data, causes  the model to forget some of  its initial skills. This effect is called “catastrophic forgetting” and has been a known issue since the first release of the foundation language models or \"LMs.\"\n\n\"Data Drama!\" – Tackling Noisy, Templated, & Boring Domain Data\nCustomer domain data is often plagu"},{"ref":"P21","kind":"page","title":"Meet Arcee Supernova Our Flagship 70b Model Alternative To Openai","date":"2026-06-27T20:01:02.499289+00:00","date_source":null,"source_url":"https://www.arcee.ai/blog/meet-arcee-supernova-our-flagship-70b-model-alternative-to-openai","signal_url":null,"signal_json_url":null,"text":"Arcee AI | Meet Arcee-SuperNova: Our Flagship 70B Model, Alternative to OpenAI \n\nTrinity Large Thinking: Available on OpenRouter.\n\nTry now ↗\n\nENTERPRISE\n\nResearch\n\nCOMPANY\n\nGet API\n\nBlog\n/\nMeet Arcee-SuperNova: Our Flagship 70B Model, Alternative to OpenAI\n\nMeet Arcee-SuperNova: Our Flagship 70B Model, Alternative to OpenAI\nLucas Atkins\n,\n\nFernando Fernandes\n,\n\n•\n\nSeptember 10, 2024\n\nMeet Arcee-SuperNova: a groundbreaking model with state-of-the-art abilities in instruction-following and strong alignment with human preferences.\n\nThis report outlines the development, performance, and deployment strategies of Arcee-SuperNova , our latest model release, which serves as the flagship general model for our OpenAI Migration plan. This model represents a significant evolution in our approach to large language models, particularly in terms of instruction-following capabilities, alignment with human preferences, and customer integration.\nModel Composition and Post-Training Techniques\nArcee-SuperNova is the result of integrating new post-training techniques developed in-house. Specifically, the model is a distilled version of Llama-3.1-405B-Instruct into Llama-3.1-70B-Instruct , which serves as the foundation through our DistillKit . We utilized the logits and attention masks from the 405B model, preserving its instruction-following capabilities while reducing model size.\nIn parallel, we trained a separate model based on base Llama-3.1-70B, using synthetic instruction data generated through our Evol-Kit pipeline. This pipeline played a pivotal role in enhancing the model’s ability to respond to diverse queries with precision and strict adherence to user instructions. To further optimize the integration of updates during training, we adopt a merging technique every half epoch, enabling smoother and more consistent performance improvements.\nA third version of Llama-3.1-70B-Instruct was also trained with additional Direct Preference Optimization (DPO) , aimed at refining the model&#x27;s alignment with human preferences. Although this DPO-enhanced version was included in the final model merge , it contributed a lesser total weight compared to the distilled and instruction-tun"},{"ref":"P22","kind":"page","title":"Distilling Kimi Delta Attention Into Afm 4 5b And The Tool We Used To Do It","date":"2026-06-27T20:01:02.42512+00:00","date_source":null,"source_url":"https://www.arcee.ai/blog/distilling-kimi-delta-attention-into-afm-4-5b-and-the-tool-we-used-to-do-it","signal_url":null,"signal_json_url":null,"text":"Arcee AI | Distilling Kimi Delta Attention into AFM-4.5B \n\nTrinity Large Thinking: Available on OpenRouter.\n\nTry now ↗\n\nENTERPRISE\n\nResearch\n\nCOMPANY\n\nGet API\n\nBlog\n/\nDistilling Kimi Delta Attention into AFM-4.5B (and the Tool We Used to Do It)\n\nDistilling Kimi Delta Attention into AFM-4.5B (and the Tool We Used to Do It)\nCharles Goddard\n,\n\n•\n\nDecember 15, 2025\n\nLearn how Kimi Delta Attention was distilled into AFM-4.5B using knowledge distillation, long-context training, and Arcee’s open-source DistillKit.\n\nMoonshot AI recently put out a great paper (and an associated model ) on an extension of Gated DeltaNet that they have termed Kimi Delta Attention (KDA). The results look super promising, particularly in the now-classic three-to-one interleaved local and global attention hybrid arrangement. The pretrained model they released is great, but they&#x27;ve open sourced both training and inference kernels, so of course I had to do something to play with them.\nPretraining a whole model from scratch just for funsies is still a little too rich for my blood. Inspired by the paper RADLADS , I decided to try to convert AFM-4.5B-Base into a hybrid KDA and full-attention transformer through knowledge distillation, then see how far it generalizes in long context land.\nA tiny note on terms before we go too far: when I say “full attention” I mean standard global self-attention. When I say “NoPE” in this post, I mean we removed RoPE and did not replace it with any other positional embedding scheme in those layers.\nCreating the Student\nFirst order of business was to create the student model, meaning both modeling code for the desired architecture and a set of decently-initialized weights.\nModeling code turned out to be super easy thanks to flash-linear-attention . The Moonshot AI folks contributed kernels and there&#x27;s a complete layer implementation that is more or less a drop-in fit. The only real code changes needed were to plug that in, rip out RoPE, and add configuration for what layers are KDA vs. full attention.\nInitializing weights is a little trickier, but not much.\nObviously for the majority of the weights they can be copied straight through from the teacher to th"},{"ref":"P23","kind":"page","title":"Small Models Big Impact How Arcee Ai Is Redefining Ai For Enterprises","date":"2026-06-27T20:01:02.264415+00:00","date_source":null,"source_url":"https://www.arcee.ai/blog/small-models-big-impact-how-arcee-ai-is-redefining-ai-for-enterprises","signal_url":null,"signal_json_url":null,"text":"Arcee AI | Small Models, Big Impact: How Arcee Is Redefining AI for Enterprises \n\nTrinity Large Thinking: Available on OpenRouter.\n\nTry now ↗\n\nENTERPRISE\n\nResearch\n\nCOMPANY\n\nGet API\n\nBlog\n/\nSmall Models, Big Impact: How Arcee AI Is Redefining AI for Enterprises\n\nSmall Models, Big Impact: How Arcee AI Is Redefining AI for Enterprises\nGrit Daily\n,\n\n•\n\nJanuary 10, 2025\n\nThe AI landscape is crowded with ambitious startups, each claiming to revolutionize the way we live, work, or analyze data. Yet, amid the buzzwords and\n\nThe AI landscape is crowded with ambitious startups, each claiming to revolutionize the way we live, work, or analyze data. Yet, amid the buzzwords and rapid-fire innovations, Arcee is carving out a unique space. Helmed by CEO Mark McQuade , the company is proving that smaller, specialized AI models can deliver outsized value, particularly for enterprises seeking cost-effective, scalable solutions that align with their business goals.\n\nSmaller Models, Greater Control\nArcee isn’t chasing the size-obsessed race of large language models (LLMs). Instead, the company focuses on smaller, specialized models that offer businesses a distinct advantage: control. These models can run on proprietary hardware, in private clouds, or on Arcee’s SaaS platform, giving enterprises full ownership of their data and the model itself.\nMcQuade emphasizes the importance of this ownership. Startups and enterprises alike benefit from owning their AI assets, and not just when it comes to cutting costs. It is about building defensible, scalable solutions that investors take seriously.\nThis approach stands in contrast to widely used public APIs like ChatGPT, which McQuade sees as effective but potentially limiting for businesses with high-scale needs or unique use cases.\nDistillation and Post-Training\nAt the core of Arcee’s innovation is its sophisticated post-training pipeline. Using proprietary tools like DistillKit and MergeKit, the company distills massive models into smaller, task-specific ones without sacrificing performance. For instance, McQuade shared how the team recently reduced a 670-billion-parameter model into something manageable for everyday enterprise use.\n“We "},{"ref":"P24","kind":"page","title":"The Case For Small Language Model Inference On Arm Cpus","date":"2026-06-27T20:01:02.235045+00:00","date_source":null,"source_url":"https://www.arcee.ai/blog/the-case-for-small-language-model-inference-on-arm-cpus","signal_url":null,"signal_json_url":null,"text":"Arcee AI | The Case for Small Language Model Inference on Arm CPUs \n\nTrinity Large Thinking: Available on OpenRouter.\n\nTry now ↗\n\nENTERPRISE\n\nResearch\n\nCOMPANY\n\nGet API\n\nBlog\n/\nThe Case for Small Language Model Inference on Arm CPUs\n\nThe Case for Small Language Model Inference on Arm CPUs\nJulien Simon\n,\n\n•\n\nApril 17, 2025\n\nOur Chief Evangelist, Julien Simon, explores the advantages and practical applications of running SLM inference on Arm CPUs.\n\nIn the dynamic realm of Artificial Intelligence (AI), Small Language Models (SLMs) are emerging as indispensable tools for organizations. Their unique blend of performance, cost-effectiveness, and resource efficiency is reshaping the AI landscape. As the demand for AI-driven solutions escalates across industries,SLMs present a compelling inference scenario on a variety of hardware, including Arm CPUs. This blog post delves into the advantages and practical applications of running SLM inference on Arm CPUs, underscoring how high-efficiency cloud architectures based on Arm CPUs are set to redefine the reach and cost-effectiveness of AI solutions.\nFrom LLMs to SLMs\n\nIn the past few years, we have witnessed a significant shift in the AI landscape, marked by the rise of large language models (LLMs) that have showcased impressive capabilities in natural language understanding and generation. However, the sheer size and computational demands of these models often render them impractical for many real-world applications. This is where the more compact and efficient small language models step in, maintaining high levels of accuracy while being more practical. Recent advancements in model architecture and optimization techniques, such as knowledge distillation, have made it possible for Virtuoso-Lite , a 10-billion parameter SLM recently released by Arcee AI , to outperform Nova , a 70-billion parameter model also released by Arcee AI in July 2024 and the best open-source model in its size range at the time. This shift toward smaller models is not just about reducing model size; it&#x27;s about making best-in-class models accessible across a wide range of environments, from edge devices to cloud servers, without the need for expe"},{"ref":"P25","kind":"page","title":"Trinity Large","date":"2026-06-27T20:01:02.024249+00:00","date_source":null,"source_url":"https://www.arcee.ai/blog/trinity-large","signal_url":null,"signal_json_url":null,"text":"Arcee AI | Trinity Large: An Open 400B Sparse MoE Model \n\nTrinity Large Thinking: Available on OpenRouter.\n\nTry now ↗\n\nENTERPRISE\n\nResearch\n\nCOMPANY\n\nGet API\n\nBlog\n/\nTrinity Large\n\nTrinity Large\nLucas Atkins\n,\n\n•\n\nJanuary 27, 2026\n\nA deep dive into Trinity Large, covering architecture, sparsity, training at scale, and why we shipped Preview, Base, and TrueBase checkpoints.\n\nTwo months ago I wrote about why we decided to stop treating pretraining like someone else&#x27;s job.\nAt the time, Trinity Nano Preview and Trinity Mini had just released, and Trinity Large had started training. We were in the middle of our first run so big that you either laughed or got nauseous. Frankly, I felt either we’d end up with a really great base model or fall flat on our faces with a tired wallet.\nLittle did I know, we’d get both.\nHere’s what we’re shipping, what surprised us, what broke, and what it took to make a 400B sparse MoE behave.\nWe&#x27;re putting out three variants: Trinity-Large- Preview is lightly post-trained and chat-ready, Trinity-Large- Base is our best pretraining checkpoint after the full 17T recipe, and TrueBase is an early checkpoint from the same run at 10T tokens, without any instruct data or LR anneals. What many would consider a true base model.\nTrinity-Large is a 400B parameter sparse MoE with 13B active parameters per token . It uses 256 experts with 4 experts active per token . That sparsity ratio is pretty high compared to our peers, save for Llama-4-Maverick:\n\nModel \nRouting (k-of-N) \nRouting fraction \n\nTrinity Large \n4-of-256 \n1.56% \n\nDeepSeek-V3 \n8-of-256 \n3.13% \n\nMiniMax-M2 \n8-of-256 \n3.13% \n\nGLM-4.5 \n8-of-160 \n5.0% \n\nQwen3-235B-A22B \n8-of-128 \n6.25% \n\nLlama 4 Maverick \n1-of-128 \n0.78% \n\nWe originally aimed for a slightly different total size (420B), but we ended up increasing the number of dense layers (from 3 to 6) to help keep routing stable at this sparsity.\n\nTrinity-Large-Base is a true frontier-class foundation model. We match and exceed our peers in open-base models across a wide range of benchmarks, including math, coding, scientific reasoning, and raw knowledge absorption.\nInference efficiency\nWe trained on 2048 Nvidia B300 GPUs. As far as"},{"ref":"P26","kind":"page","title":"Virtuoso Lite Virtuoso Medium V2 Distilling Deepseek V3 Into 10b 32b Small Language Models Slms","date":"2026-06-27T20:01:01.869534+00:00","date_source":null,"source_url":"https://www.arcee.ai/blog/virtuoso-lite-virtuoso-medium-v2-distilling-deepseek-v3-into-10b-32b-small-language-models-slms","signal_url":null,"signal_json_url":null,"text":"Arcee AI | Virtuoso-Lite & Virtuoso-Medium-v2: Distilling DeepSeek-V3 into 10B & 32B Small Language Models (SLMs) \n\nTrinity Large Thinking: Available on OpenRouter.\n\nTry now ↗\n\nENTERPRISE\n\nResearch\n\nCOMPANY\n\nGet API\n\nBlog\n/\nVirtuoso-Lite & Virtuoso-Medium-v2: Distilling DeepSeek-V3 into 10B & 32B Small Language Models (SLMs)\n\nVirtuoso-Lite & Virtuoso-Medium-v2: Distilling DeepSeek-V3 into 10B & 32B Small Language Models (SLMs)\nLucas Atkins\n,\n\n•\n\nJanuary 28, 2025\n\nHere at Arcee AI, we&#x27;re going beyond the hype and speculation surrounding the DeepSeek R-1 release. We&#x27;re doing what we always do: work hard on training models. Soon we will deliver you disillations of R-1, and in the meantime, we&#x27;re bringing you two distilled versions (10B, 32B) of DeepSeek-V3.\n\nOver the past few days, the reactions to DeepSeek-R1 have been dramatic—with some voices welcoming it, others resisting it, and many voices speculating wildly. \nAt Arcee AI, we’re not here to debate those reactions; we’re here to deliver something tangible. \nToday, we’re excited to share two new small language models (SLMs) we’ve been working on that reflect our core commitment: turning top-tier research into reliable, effective AI tools for everyone. Today’s releases are both distilled versions of DeepSeek V3, and we’ll be sharing DeepSeek-R1 distillations coming very soon.\nIntroducing Virtuoso-Lite (10B)\n  Virtuoso-Lite is a brand-new 10B-parameter model built on top of TII’s Falcon architecture. We took the following steps to bring Virtuoso-Lite to life:\nTokenizer Work: We retrained on the DeepSeek-V3 tokenizer, capturing the crucial distribution of tokens needed for effective cross-architecture distillation.\nDistillation with fp8 DeepSeek-V3: We distilled 1.1 billion tokens’ worth of raw output values, also known as logits, from an 8-bit version of DeepSeek-V3 670B to preserve the “essence” of performance while significantly reducing the parameter count.\nTokenizer Reversion & Merging: After the initial distillation, we returned to a Llama-3 tokenizer, merged everything, and applied Reinforcement Learning with Human Feedback (RLHF) to fine-tune and align the final model.\n\nThe result is a strea"},{"ref":"P27","kind":"page","title":"Trinity Large Thinking","date":"2026-06-27T20:01:01.847067+00:00","date_source":null,"source_url":"https://www.arcee.ai/blog/trinity-large-thinking","signal_url":null,"signal_json_url":null,"text":"Arcee AI | Trinity-Large-Thinking: Scaling an Open Source Frontier Agent \n\nTrinity Large Thinking: Available on OpenRouter.\n\nTry now ↗\n\nENTERPRISE\n\nResearch\n\nCOMPANY\n\nGet API\n\nBlog\n/\nTrinity-Large-Thinking: Scaling an Open Source Frontier Agent\n\nTrinity-Large-Thinking: Scaling an Open Source Frontier Agent\nLucas Atkins\n,\n\n•\n\nApril 1, 2026\n\nTrinity-Large-Thinking is live. A frontier open reasoning model for complex, long-horizon agents and multi-turn tool calling released under Apache 2.0.\n\nToday we are releasing Trinity-Large-Thinking on our API and the weights on Hugging Face under the Apache 2.0 license.\nNine months ago, we made a decision that changed the shape of this company. We decided that if we cared about serious American open models, about models developers and enterprises could actually own, then we needed to build them ourselves.\nThat decision gave us Trinity.\nFirst came the smaller models, 4.5B, Nano and Mini. Then came Preview at the end of January, our first public look at Trinity Large.\nToday comes the official release: Trinity-Large-Thinking, our reasoning model built to close the gaps Preview left open.\nOn many axes, it is the strongest open model ever released outside of China.\nIt is the result of the last two months spent improving and scaling our SFT and RL pipeline so it could meet the size and capability of the Trinity-Large base model. Preview was an instruct model, as the name would suggest, the new checkpoint uses “thinking” prior to responding, like Trinity-Mini does. This enables stronger multi-turn tool calling, better context coherence, cleaner instruction following, and more stable behavior across long-running agent loops.\n\nThe market has been asking for this. Trinity-Large-Preview took off faster than we expected.\nWe launched it at the end of January as a light instruct post-train, with the expectation that people would test it, break it, and help show us where it wanted to go next. That is exactly what happened. The model later crossed just 3.37 trillion tokens served on OpenRouter in its first 2 months, and OpenRouter’s OpenClaw collection has Trinity Large Preview as the #1 most used open model in the U.S. and #4 globally.\n\nAnd"},{"ref":"P28","kind":"page","title":"Arcee Ai Small Language Models On Together Ai And Openrouter","date":"2026-06-27T20:01:01.530136+00:00","date_source":null,"source_url":"https://www.arcee.ai/blog/arcee-ai-small-language-models-on-together-ai-and-openrouter","signal_url":null,"signal_json_url":null,"text":"Arcee AI | Arcee AI Small Language Models Now Available on Together.ai and OpenRouter \n\nTrinity Large Thinking: Available on OpenRouter.\n\nTry now ↗\n\nENTERPRISE\n\nResearch\n\nCOMPANY\n\nGet API\n\nBlog\n/\nArcee AI SLMs Now Available on Together.ai and OpenRouter\n\nArcee AI SLMs Now Available on Together.ai and OpenRouter\nJulien Simon\n,\n\n•\n\nMay 27, 2025\n\nArcee AI is excited to announce the availability of its small language models (SLMs) on Together.ai and OpenRouter, two leading managed inference platforms. Start building today and leverage Arcee AI’s specialized models to enhance your AI applications.\n\nToday, we are thrilled to announce that our suite of small language models (SLMs) is now available on Together.ai and OpenRouter , two leading platforms for managed, pay-per-token inference services. With this seamless integration, you can start building high-quality, fast, and cost-efficient AI-powered applications in minutes, without having to manage complex AI infrastructure. \nWhether you’re looking for models for general-purpose chat applications, code generation, function calling, reasoning, or image-to-text, we’ve got you covered! Here are the available models.\n\nModel Name \nDescription \nSize \nContext Window \nPrice per Million Input Tokens \nPrice per Million Output Tokens \nTogether.ai \nOpenRouter \n\nArcee Blitz \nFast, cost-efficient model for everyday chat. \n24B \n33K \n$0.45 \n$0.75 \nLink \nLink \n\nVirtuoso Medium V2 \nBalanced model for a wide range of tasks. \n32B \n131K \n$0.50 \n$0.80 \nLink \nLink \n\nVirtuoso Large \nGeneral-purpose LLM for cross-domain reasoning and creative writing. \n72B \n131K \n$0.75 \n$1.20 \nLink \nLink \n\nCoder Large \nOptimized for code generation and refactoring. \n32B \n33K \n$0.50 \n$0.80 \nLink \nLink \n\nCaller Large \nSpecialized for function calling and orchestration. \n33B \n33K \n$0.55 \n$0.85 \nLink \nLink \n\nMaestro Reasoning \nFlagship model for step-by-step logical reasoning. \n32B \n131K \n$0.90 \n$3.30 \nLink \nLink \n\nSpotlight \nLightning-fast vision-language model for image-text tasks. \n7B \n131K \n$0.18 \n$0.18 \nLink \nLink \n\nWhy Build with Small Language Models\nUse The Right Model for Each Task\nIn the world of AI, one size does not fit all. While frontier-scale \"Swiss"},{"ref":"E1","kind":"event","title":"arcee-ai/Trinity-Mini","date":"2025-12-01T16:12:10+00:00","date_source":"source","source_url":"https://huggingface.co/arcee-ai/Trinity-Mini","signal_url":"https://onlylabs.fyi/signals/5b0a8b46-a677-408c-aa20-018d31cecad2","signal_json_url":"https://onlylabs.fyi/signals/5b0a8b46-a677-408c-aa20-018d31cecad2/signal.json","text":"model_released · arcee-ai/Trinity-Mini · signal_desk=releases · occurred_at=2025-12-01T16:12:10+00:00 · url=https://huggingface.co/arcee-ai/Trinity-Mini · hf_downloads=25076 · hf_likes=193 · hf_params=26123974400 · pipeline=text-generation · license=other · raw={\"derived_reason\":\"first-party-finetune\"}"},{"ref":"E2","kind":"event","title":"arcee-ai/Trinity-Large-Thinking","date":"2026-04-01T03:35:06+00:00","date_source":"source","source_url":"https://huggingface.co/arcee-ai/Trinity-Large-Thinking","signal_url":"https://onlylabs.fyi/signals/429af5af-ee30-4a25-874a-139c404e5dec","signal_json_url":"https://onlylabs.fyi/signals/429af5af-ee30-4a25-874a-139c404e5dec/signal.json","text":"model_released · arcee-ai/Trinity-Large-Thinking · signal_desk=releases · occurred_at=2026-04-01T03:35:06+00:00 · url=https://huggingface.co/arcee-ai/Trinity-Large-Thinking · hf_downloads=8014 · hf_likes=183 · hf_params=398635286016 · pipeline=text-generation · license=other · raw={\"derived_reason\":\"first-party-finetune\"}"},{"ref":"E3","kind":"event","title":"arcee-ai/Trinity-Large-Preview","date":"2026-01-27T14:52:00+00:00","date_source":"source","source_url":"https://huggingface.co/arcee-ai/Trinity-Large-Preview","signal_url":"https://onlylabs.fyi/signals/b357f482-7717-4ab4-ae03-f146b4efef08","signal_json_url":"https://onlylabs.fyi/signals/b357f482-7717-4ab4-ae03-f146b4efef08/signal.json","text":"model_released · arcee-ai/Trinity-Large-Preview · signal_desk=releases · occurred_at=2026-01-27T14:52:00+00:00 · url=https://huggingface.co/arcee-ai/Trinity-Large-Preview · hf_downloads=734 · hf_likes=180 · hf_params=398635286016 · pipeline=text-generation · license=other · raw={\"derived_reason\":\"first-party-finetune\"}"},{"ref":"E4","kind":"event","title":"arcee-ai/Homunculus","date":"2025-06-02T18:46:57+00:00","date_source":"source","source_url":"https://huggingface.co/arcee-ai/Homunculus","signal_url":"https://onlylabs.fyi/signals/41f75527-4bba-4abe-85bb-afeaf74bf2bd","signal_json_url":"https://onlylabs.fyi/signals/41f75527-4bba-4abe-85bb-afeaf74bf2bd/signal.json","text":"model_released · arcee-ai/Homunculus · signal_desk=releases · occurred_at=2025-06-02T18:46:57+00:00 · url=https://huggingface.co/arcee-ai/Homunculus · hf_downloads=15 · hf_likes=99 · hf_params=12458808320 · pipeline=text-generation · license=apache-2.0 · raw={\"derived_reason\":\"first-party-finetune\"}"},{"ref":"E5","kind":"event","title":"arcee-ai/AFM-4.5B","date":"2025-07-29T15:04:18+00:00","date_source":"source","source_url":"https://huggingface.co/arcee-ai/AFM-4.5B","signal_url":"https://onlylabs.fyi/signals/cca31fed-0731-4479-8109-726c875f6af0","signal_json_url":"https://onlylabs.fyi/signals/cca31fed-0731-4479-8109-726c875f6af0/signal.json","text":"model_released · arcee-ai/AFM-4.5B · signal_desk=releases · occurred_at=2025-07-29T15:04:18+00:00 · url=https://huggingface.co/arcee-ai/AFM-4.5B · hf_downloads=6253 · hf_likes=96 · hf_params=4619189760 · pipeline=text-generation · license=apache-2.0 · raw={\"derived_reason\":\"first-party-finetune\"}"},{"ref":"E6","kind":"event","title":"arcee-ai/Trinity-Nano-Preview","date":"2025-12-01T19:16:24+00:00","date_source":"source","source_url":"https://huggingface.co/arcee-ai/Trinity-Nano-Preview","signal_url":"https://onlylabs.fyi/signals/f8a9e0d5-b686-4259-84b2-13c5d5c1f48e","signal_json_url":"https://onlylabs.fyi/signals/f8a9e0d5-b686-4259-84b2-13c5d5c1f48e/signal.json","text":"model_released · arcee-ai/Trinity-Nano-Preview · signal_desk=releases · occurred_at=2025-12-01T19:16:24+00:00 · url=https://huggingface.co/arcee-ai/Trinity-Nano-Preview · hf_downloads=23994 · hf_likes=79 · hf_params=6120003328 · pipeline=text-generation · license=other · raw={\"derived_reason\":\"first-party-finetune\"}"},{"ref":"E7","kind":"event","title":"arcee-ai/Arcee-Blitz","date":"2025-02-07T07:55:43+00:00","date_source":"source","source_url":"https://huggingface.co/arcee-ai/Arcee-Blitz","signal_url":"https://onlylabs.fyi/signals/5af8cd6f-d935-4816-9d1d-fbb1a85133dc","signal_json_url":"https://onlylabs.fyi/signals/5af8cd6f-d935-4816-9d1d-fbb1a85133dc/signal.json","text":"model_released · arcee-ai/Arcee-Blitz · signal_desk=releases · occurred_at=2025-02-07T07:55:43+00:00 · url=https://huggingface.co/arcee-ai/Arcee-Blitz · hf_downloads=37 · hf_likes=79 · hf_params=23572403200 · pipeline=text-generation · license=apache-2.0 · raw={\"derived_reason\":\"first-party-finetune\"}"},{"ref":"E8","kind":"event","title":"arcee-ai/Trinity-Large-TrueBase","date":"2026-01-27T10:15:24+00:00","date_source":"source","source_url":"https://huggingface.co/arcee-ai/Trinity-Large-TrueBase","signal_url":"https://onlylabs.fyi/signals/a4e0c878-5f8b-40c6-9f98-4ee804203235","signal_json_url":"https://onlylabs.fyi/signals/a4e0c878-5f8b-40c6-9f98-4ee804203235/signal.json","text":"model_released · arcee-ai/Trinity-Large-TrueBase · signal_desk=releases · occurred_at=2026-01-27T10:15:24+00:00 · url=https://huggingface.co/arcee-ai/Trinity-Large-TrueBase · hf_downloads=290 · hf_likes=67 · hf_params=398635286016 · pipeline=text-generation · license=other"},{"ref":"E9","kind":"event","title":"arcee-ai/Trinity-Large-Base","date":"2026-01-27T11:31:37+00:00","date_source":"source","source_url":"https://huggingface.co/arcee-ai/Trinity-Large-Base","signal_url":"https://onlylabs.fyi/signals/d077f164-5763-41f9-925d-f7baa485ef29","signal_json_url":"https://onlylabs.fyi/signals/d077f164-5763-41f9-925d-f7baa485ef29/signal.json","text":"model_released · arcee-ai/Trinity-Large-Base · signal_desk=releases · occurred_at=2026-01-27T11:31:37+00:00 · url=https://huggingface.co/arcee-ai/Trinity-Large-Base · hf_downloads=231 · hf_likes=58 · hf_params=398635286016 · pipeline=text-generation · license=other · raw={\"derived_reason\":\"first-party-finetune\"}"},{"ref":"E10","kind":"event","title":"arcee-ai/Arcee-Maestro-7B-Preview","date":"2025-02-10T23:23:19+00:00","date_source":"source","source_url":"https://huggingface.co/arcee-ai/Arcee-Maestro-7B-Preview","signal_url":"https://onlylabs.fyi/signals/b1f3057c-b69a-4bba-9782-91814d66b9bd","signal_json_url":"https://onlylabs.fyi/signals/b1f3057c-b69a-4bba-9782-91814d66b9bd/signal.json","text":"model_released · arcee-ai/Arcee-Maestro-7B-Preview · signal_desk=releases · occurred_at=2025-02-10T23:23:19+00:00 · url=https://huggingface.co/arcee-ai/Arcee-Maestro-7B-Preview · hf_downloads=98 · hf_likes=38 · hf_params=7612756480 · pipeline=text-generation · license=apache-2.0 · raw={\"derived_reason\":\"first-party-finetune\"}"},{"ref":"E11","kind":"event","title":"arcee-ai/GLM-4-32B-Base-32K","date":"2025-06-23T14:29:20+00:00","date_source":"source","source_url":"https://huggingface.co/arcee-ai/GLM-4-32B-Base-32K","signal_url":"https://onlylabs.fyi/signals/c525a4e1-e143-4008-83f5-8f63e0d71874","signal_json_url":"https://onlylabs.fyi/signals/c525a4e1-e143-4008-83f5-8f63e0d71874/signal.json","text":"model_released · arcee-ai/GLM-4-32B-Base-32K · signal_desk=releases · occurred_at=2025-06-23T14:29:20+00:00 · url=https://huggingface.co/arcee-ai/GLM-4-32B-Base-32K · hf_downloads=30 · hf_likes=38 · hf_params=32563513344 · pipeline=text-generation · license=mit · raw={\"derived_reason\":\"first-party-finetune\"}"},{"ref":"E12","kind":"event","title":"arcee-ai/mergekit","date":"2023-08-21T03:50:04+00:00","date_source":"source","source_url":"https://github.com/arcee-ai/mergekit","signal_url":"https://onlylabs.fyi/signals/1304d1b2-f092-4d48-9522-ea120a0dd170","signal_json_url":"https://onlylabs.fyi/signals/1304d1b2-f092-4d48-9522-ea120a0dd170/signal.json","text":"repo_new · arcee-ai/mergekit · signal_desk=repos · occurred_at=2023-08-21T03:50:04+00:00 · url=https://github.com/arcee-ai/mergekit · stars=7186 · raw={\"repo\":\"arcee-ai/mergekit\",\"description\":\"Tools for merging pretrained large language models.\",\"language\":\"Python\"}"},{"ref":"E13","kind":"event","title":"arcee-ai/AFM-4.5B-Base","date":"2025-07-29T15:01:10+00:00","date_source":"source","source_url":"https://huggingface.co/arcee-ai/AFM-4.5B-Base","signal_url":"https://onlylabs.fyi/signals/2e6a8619-e906-45af-8cac-a777dee82a85","signal_json_url":"https://onlylabs.fyi/signals/2e6a8619-e906-45af-8cac-a777dee82a85/signal.json","text":"model_released · arcee-ai/AFM-4.5B-Base · signal_desk=releases · occurred_at=2025-07-29T15:01:10+00:00 · url=https://huggingface.co/arcee-ai/AFM-4.5B-Base · hf_downloads=29736 · hf_likes=34 · hf_params=4619184640 · pipeline=text-generation · license=apache-2.0"},{"ref":"E14","kind":"event","title":"arcee-ai/Virtuoso-Small-v2","date":"2025-01-30T05:41:29+00:00","date_source":"source","source_url":"https://huggingface.co/arcee-ai/Virtuoso-Small-v2","signal_url":"https://onlylabs.fyi/signals/65d4f427-f0c2-4648-8019-7b07f1a07e4f","signal_json_url":"https://onlylabs.fyi/signals/65d4f427-f0c2-4648-8019-7b07f1a07e4f/signal.json","text":"model_released · arcee-ai/Virtuoso-Small-v2 · signal_desk=releases · occurred_at=2025-01-30T05:41:29+00:00 · url=https://huggingface.co/arcee-ai/Virtuoso-Small-v2 · hf_downloads=1349 · hf_likes=35 · hf_params=14765947904 · pipeline=text-generation · license=apache-2.0 · raw={\"derived_reason\":\"first-party-finetune\"}"},{"ref":"E15","kind":"event","title":"arcee-ai/Virtuoso-Large","date":"2025-06-09T17:23:45+00:00","date_source":"source","source_url":"https://huggingface.co/arcee-ai/Virtuoso-Large","signal_url":"https://onlylabs.fyi/signals/3a9535a1-b0ed-4bc6-824d-ea39702635ec","signal_json_url":"https://onlylabs.fyi/signals/3a9535a1-b0ed-4bc6-824d-ea39702635ec/signal.json","text":"model_released · arcee-ai/Virtuoso-Large · signal_desk=releases · occurred_at=2025-06-09T17:23:45+00:00 · url=https://huggingface.co/arcee-ai/Virtuoso-Large · hf_downloads=32 · hf_likes=30 · hf_params=72706203648 · pipeline=text-generation · license=other · raw={\"derived_reason\":\"first-party-finetune\"}"},{"ref":"E16","kind":"event","title":"arcee-ai/Trinity-Nano-Base","date":"2025-12-01T18:28:08+00:00","date_source":"source","source_url":"https://huggingface.co/arcee-ai/Trinity-Nano-Base","signal_url":"https://onlylabs.fyi/signals/f42fb7c7-bfa3-4215-930c-a7f7f692c5be","signal_json_url":"https://onlylabs.fyi/signals/f42fb7c7-bfa3-4215-930c-a7f7f692c5be/signal.json","text":"model_released · arcee-ai/Trinity-Nano-Base · signal_desk=releases · occurred_at=2025-12-01T18:28:08+00:00 · url=https://huggingface.co/arcee-ai/Trinity-Nano-Base · hf_downloads=2526 · hf_likes=28 · hf_params=6120003328 · pipeline=text-generation · license=other · raw={\"derived_reason\":\"first-party-finetune\"}"},{"ref":"E17","kind":"event","title":"Technical AI Account Manager","date":"2026-05-22T14:15:42+00:00","date_source":"greenhouse.updated_at","source_url":"https://www.arcee.ai/careers?gh_jid=5229012008","signal_url":"https://onlylabs.fyi/signals/db9d7227-db02-4008-bed9-862d9f43f94b","signal_json_url":"https://onlylabs.fyi/signals/db9d7227-db02-4008-bed9-862d9f43f94b/signal.json","text":"job_opened · Technical AI Account Manager · signal_desk=hiring · occurred_at=2026-05-22T14:15:42+00:00 · url=https://www.arcee.ai/careers?gh_jid=5229012008 · raw={\"location\":\"SF, CA\",\"ats\":\"greenhouse\"}"},{"ref":"E18","kind":"event","title":"Compute Infrastructure Specialist","date":"2026-05-22T14:08:28+00:00","date_source":"greenhouse.updated_at","source_url":"https://www.arcee.ai/careers?gh_jid=5228121008","signal_url":"https://onlylabs.fyi/signals/5d846be1-457d-41c6-8868-b134f1ee8976","signal_json_url":"https://onlylabs.fyi/signals/5d846be1-457d-41c6-8868-b134f1ee8976/signal.json","text":"job_opened · Compute Infrastructure Specialist · signal_desk=hiring · occurred_at=2026-05-22T14:08:28+00:00 · url=https://www.arcee.ai/careers?gh_jid=5228121008 · raw={\"location\":\"SF, CA\",\"ats\":\"greenhouse\"}"},{"ref":"E19","kind":"event","title":"arcee-ai/Trinity-Mini-Base","date":"2025-12-01T18:25:42+00:00","date_source":"source","source_url":"https://huggingface.co/arcee-ai/Trinity-Mini-Base","signal_url":"https://onlylabs.fyi/signals/4287d123-891d-413e-ab05-182b0dfc4851","signal_json_url":"https://onlylabs.fyi/signals/4287d123-891d-413e-ab05-182b0dfc4851/signal.json","text":"model_released · arcee-ai/Trinity-Mini-Base · signal_desk=releases · occurred_at=2025-12-01T18:25:42+00:00 · url=https://huggingface.co/arcee-ai/Trinity-Mini-Base · hf_downloads=369 · hf_likes=22 · hf_params=26123974400 · pipeline=text-generation · license=other · raw={\"derived_reason\":\"first-party-finetune\"}"},{"ref":"E20","kind":"event","title":"arcee-ai/Arcee-SuperNova-v1","date":"2025-06-10T07:52:25+00:00","date_source":"source","source_url":"https://huggingface.co/arcee-ai/Arcee-SuperNova-v1","signal_url":"https://onlylabs.fyi/signals/3d216abc-b8d7-4742-9587-6e5bd2d41389","signal_json_url":"https://onlylabs.fyi/signals/3d216abc-b8d7-4742-9587-6e5bd2d41389/signal.json","text":"model_released · arcee-ai/Arcee-SuperNova-v1 · signal_desk=releases · occurred_at=2025-06-10T07:52:25+00:00 · url=https://huggingface.co/arcee-ai/Arcee-SuperNova-v1 · hf_downloads=20 · hf_likes=17 · hf_params=70553706496 · pipeline=text-generation · license=llama3 · raw={\"derived_reason\":\"first-party-finetune\"}"},{"ref":"E21","kind":"event","title":"arcee-ai/DALM","date":"2023-07-31T23:23:58+00:00","date_source":"source","source_url":"https://github.com/arcee-ai/DALM","signal_url":"https://onlylabs.fyi/signals/37fc1d92-f927-44e7-b257-99f2009e6899","signal_json_url":"https://onlylabs.fyi/signals/37fc1d92-f927-44e7-b257-99f2009e6899/signal.json","text":"repo_new · arcee-ai/DALM · signal_desk=repos · occurred_at=2023-07-31T23:23:58+00:00 · url=https://github.com/arcee-ai/DALM · stars=341 · hn=3 points/1 comments · raw={\"repo\":\"arcee-ai/DALM\",\"description\":\"Domain Adapted Language Modeling Toolkit - E2E RAG\",\"language\":\"Python\"}"},{"ref":"E22","kind":"event","title":"arcee-ai/Trinity-Mini-Base-Pre-Anneal","date":"2025-12-11T17:42:16+00:00","date_source":"source","source_url":"https://huggingface.co/arcee-ai/Trinity-Mini-Base-Pre-Anneal","signal_url":"https://onlylabs.fyi/signals/87f44530-ad15-4971-b256-af2eaf6a27ed","signal_json_url":"https://onlylabs.fyi/signals/87f44530-ad15-4971-b256-af2eaf6a27ed/signal.json","text":"model_released · arcee-ai/Trinity-Mini-Base-Pre-Anneal · signal_desk=releases · occurred_at=2025-12-11T17:42:16+00:00 · url=https://huggingface.co/arcee-ai/Trinity-Mini-Base-Pre-Anneal · hf_downloads=304 · hf_likes=14 · hf_params=26123974400 · pipeline=text-generation · license=other"},{"ref":"E23","kind":"event","title":"arcee-ai/Trinity-Nano-Base-Pre-Anneal","date":"2025-12-11T17:41:51+00:00","date_source":"source","source_url":"https://huggingface.co/arcee-ai/Trinity-Nano-Base-Pre-Anneal","signal_url":"https://onlylabs.fyi/signals/5ec033eb-c521-476c-aad0-fcef9f419b34","signal_json_url":"https://onlylabs.fyi/signals/5ec033eb-c521-476c-aad0-fcef9f419b34/signal.json","text":"model_released · arcee-ai/Trinity-Nano-Base-Pre-Anneal · signal_desk=releases · occurred_at=2025-12-11T17:41:51+00:00 · url=https://huggingface.co/arcee-ai/Trinity-Nano-Base-Pre-Anneal · hf_downloads=110 · hf_likes=14 · hf_params=6120003328 · pipeline=text-generation · license=other"},{"ref":"E24","kind":"event","title":"arcee-ai/AFM-4.5B-Base-KDA-NoPE","date":"2025-12-14T21:35:17+00:00","date_source":"source","source_url":"https://huggingface.co/arcee-ai/AFM-4.5B-Base-KDA-NoPE","signal_url":"https://onlylabs.fyi/signals/b8d9abb6-e894-46ef-9a30-93ce6bcf43fc","signal_json_url":"https://onlylabs.fyi/signals/b8d9abb6-e894-46ef-9a30-93ce6bcf43fc/signal.json","text":"model_released · arcee-ai/AFM-4.5B-Base-KDA-NoPE · signal_desk=releases · occurred_at=2025-12-14T21:35:17+00:00 · url=https://huggingface.co/arcee-ai/AFM-4.5B-Base-KDA-NoPE · hf_downloads=7 · hf_likes=14 · hf_params=4999169436 · pipeline=image-feature-extraction · license=apache-2.0 · raw={\"derived_reason\":\"first-party-finetune\"}"},{"ref":"E25","kind":"event","title":"arcee-ai/AFM-4.5B-Base-KDA-Only","date":"2025-12-14T21:35:04+00:00","date_source":"source","source_url":"https://huggingface.co/arcee-ai/AFM-4.5B-Base-KDA-Only","signal_url":"https://onlylabs.fyi/signals/854622d8-0334-40e4-8706-9c1a05e69792","signal_json_url":"https://onlylabs.fyi/signals/854622d8-0334-40e4-8706-9c1a05e69792/signal.json","text":"model_released · arcee-ai/AFM-4.5B-Base-KDA-Only · signal_desk=releases · occurred_at=2025-12-14T21:35:04+00:00 · url=https://huggingface.co/arcee-ai/AFM-4.5B-Base-KDA-Only · hf_downloads=23 · hf_likes=12 · hf_params=5047140048 · pipeline=image-feature-extraction · license=apache-2.0 · raw={\"derived_reason\":\"first-party-finetune\"}"},{"ref":"E26","kind":"event","title":"arcee-ai/Caller","date":"2025-06-09T20:41:09+00:00","date_source":"source","source_url":"https://huggingface.co/arcee-ai/Caller","signal_url":"https://onlylabs.fyi/signals/32d71ab5-5d2d-40ed-9857-e37fa73ffae0","signal_json_url":"https://onlylabs.fyi/signals/32d71ab5-5d2d-40ed-9857-e37fa73ffae0/signal.json","text":"model_released · arcee-ai/Caller · signal_desk=releases · occurred_at=2025-06-09T20:41:09+00:00 · url=https://huggingface.co/arcee-ai/Caller · hf_downloads=10 · hf_likes=12 · hf_params=32763876352 · license=apache-2.0 · raw={\"derived_reason\":\"first-party-finetune\"}"},{"ref":"E27","kind":"event","title":"arcee-ai/AFM-4.5B-ov","date":"2025-09-05T09:00:52+00:00","date_source":"source","source_url":"https://huggingface.co/arcee-ai/AFM-4.5B-ov","signal_url":"https://onlylabs.fyi/signals/eb801ada-02cc-44f6-a8e2-69a0a23cb22c","signal_json_url":"https://onlylabs.fyi/signals/eb801ada-02cc-44f6-a8e2-69a0a23cb22c/signal.json","text":"model_released · arcee-ai/AFM-4.5B-ov · signal_desk=releases · occurred_at=2025-09-05T09:00:52+00:00 · url=https://huggingface.co/arcee-ai/AFM-4.5B-ov · hf_likes=8 · pipeline=text-generation · license=apache-2.0 · raw={\"derived_reason\":\"first-party-finetune\"}"},{"ref":"E28","kind":"event","title":"arcee-ai/Trinity-Tokenizer","date":"2025-10-08T19:21:56+00:00","date_source":"source","source_url":"https://huggingface.co/arcee-ai/Trinity-Tokenizer","signal_url":"https://onlylabs.fyi/signals/0dbd4135-8570-4464-b3ac-0d956a9ac725","signal_json_url":"https://onlylabs.fyi/signals/0dbd4135-8570-4464-b3ac-0d956a9ac725/signal.json","text":"model_released · arcee-ai/Trinity-Tokenizer · signal_desk=releases · occurred_at=2025-10-08T19:21:56+00:00 · url=https://huggingface.co/arcee-ai/Trinity-Tokenizer · hf_downloads=49 · hf_likes=6 · license=other"},{"ref":"E29","kind":"event","title":"arcee-ai/AFM-4.5B-Preview","date":"2025-06-19T16:36:25+00:00","date_source":"source","source_url":"https://huggingface.co/arcee-ai/AFM-4.5B-Preview","signal_url":"https://onlylabs.fyi/signals/c7dfe681-271c-49dd-b3dc-1916fb395e66","signal_json_url":"https://onlylabs.fyi/signals/c7dfe681-271c-49dd-b3dc-1916fb395e66/signal.json","text":"model_released · arcee-ai/AFM-4.5B-Preview · signal_desk=releases · occurred_at=2025-06-19T16:36:25+00:00 · url=https://huggingface.co/arcee-ai/AFM-4.5B-Preview · hf_downloads=9 · hf_likes=5 · hf_params=4619491840 · pipeline=text-generation · license=apache-2.0 · raw={\"derived_reason\":\"first-party-finetune\"}"},{"ref":"E30","kind":"event","title":"arcee-ai/DistillKit","date":"2024-07-31T22:16:09+00:00","date_source":"source","source_url":"https://github.com/arcee-ai/DistillKit","signal_url":"https://onlylabs.fyi/signals/a408cea1-79ee-4127-900a-ebeaad43a8ac","signal_json_url":"https://onlylabs.fyi/signals/a408cea1-79ee-4127-900a-ebeaad43a8ac/signal.json","text":"repo_new · arcee-ai/DistillKit · signal_desk=repos · occurred_at=2024-07-31T22:16:09+00:00 · url=https://github.com/arcee-ai/DistillKit · stars=973 · raw={\"repo\":\"arcee-ai/DistillKit\",\"description\":\"An Open Source Toolkit For LLM Distillation\",\"language\":\"Python\"}"},{"ref":"E31","kind":"event","title":"arcee-ai/AFM-4.5B-Base-Pre-Anneal","date":"2025-05-27T19:06:11+00:00","date_source":"source","source_url":"https://huggingface.co/arcee-ai/AFM-4.5B-Base-Pre-Anneal","signal_url":"https://onlylabs.fyi/signals/f34d3cab-c773-4142-b49f-64412e91c5f9","signal_json_url":"https://onlylabs.fyi/signals/f34d3cab-c773-4142-b49f-64412e91c5f9/signal.json","text":"model_released · arcee-ai/AFM-4.5B-Base-Pre-Anneal · signal_desk=releases · occurred_at=2025-05-27T19:06:11+00:00 · url=https://huggingface.co/arcee-ai/AFM-4.5B-Base-Pre-Anneal · hf_downloads=61 · hf_likes=3 · hf_params=4620474880 · pipeline=text-generation · license=apache-2.0"},{"ref":"E32","kind":"event","title":"arcee-ai/DeepSeek-V3-0324-bf16","date":"2025-03-24T15:18:33+00:00","date_source":"source","source_url":"https://huggingface.co/arcee-ai/DeepSeek-V3-0324-bf16","signal_url":"https://onlylabs.fyi/signals/02c5195e-6add-43c5-969a-2d3b9d01b62a","signal_json_url":"https://onlylabs.fyi/signals/02c5195e-6add-43c5-969a-2d3b9d01b62a/signal.json","text":"model_released · arcee-ai/DeepSeek-V3-0324-bf16 · signal_desk=releases · occurred_at=2025-03-24T15:18:33+00:00 · url=https://huggingface.co/arcee-ai/DeepSeek-V3-0324-bf16 · hf_downloads=10 · hf_likes=3 · hf_params=684489845504 · pipeline=text-generation · license=mit"},{"ref":"E33","kind":"event","title":"arcee-ai/fastmlx","date":"2024-03-29T18:33:55+00:00","date_source":"source","source_url":"https://github.com/arcee-ai/fastmlx","signal_url":"https://onlylabs.fyi/signals/80898ce5-e398-461c-87f6-f0f028b1c078","signal_json_url":"https://onlylabs.fyi/signals/80898ce5-e398-461c-87f6-f0f028b1c078/signal.json","text":"repo_new · arcee-ai/fastmlx · signal_desk=repos · occurred_at=2024-03-29T18:33:55+00:00 · url=https://github.com/arcee-ai/fastmlx · stars=359 · raw={\"repo\":\"arcee-ai/fastmlx\",\"description\":\"FastMLX is a high performance production ready API to host MLX models.\",\"language\":\"Python\"}"},{"ref":"E34","kind":"event","title":"arcee-ai/PruneMe","date":"2024-04-10T22:34:25+00:00","date_source":"source","source_url":"https://github.com/arcee-ai/PruneMe","signal_url":"https://onlylabs.fyi/signals/889a944d-df4e-400f-8727-c9adee77555c","signal_json_url":"https://onlylabs.fyi/signals/889a944d-df4e-400f-8727-c9adee77555c/signal.json","text":"repo_new · arcee-ai/PruneMe · signal_desk=repos · occurred_at=2024-04-10T22:34:25+00:00 · url=https://github.com/arcee-ai/PruneMe · stars=267 · raw={\"repo\":\"arcee-ai/PruneMe\",\"description\":\"Automated Identification of Redundant Layer Blocks for Pruning in Large Language Models\",\"language\":\"Python\"}"},{"ref":"E35","kind":"event","title":"arcee-ai/EvolKit","date":"2024-08-07T11:04:05+00:00","date_source":"source","source_url":"https://github.com/arcee-ai/EvolKit","signal_url":"https://onlylabs.fyi/signals/e82b6c94-1f4b-493f-83de-236750989554","signal_json_url":"https://onlylabs.fyi/signals/e82b6c94-1f4b-493f-83de-236750989554/signal.json","text":"repo_new · arcee-ai/EvolKit · signal_desk=repos · occurred_at=2024-08-07T11:04:05+00:00 · url=https://github.com/arcee-ai/EvolKit · stars=257 · raw={\"repo\":\"arcee-ai/EvolKit\",\"description\":\"EvolKit is an innovative framework designed to automatically enhance the complexity of instructions used for fine-tuning Large Language Models (LLMs). \",\"language\":\"Jupyter Notebook\"}"},{"ref":"E36","kind":"event","title":"arcee-ai/arcee-blitz-caller-beta","date":"2025-02-21T06:53:45+00:00","date_source":"source","source_url":"https://huggingface.co/arcee-ai/arcee-blitz-caller-beta","signal_url":"https://onlylabs.fyi/signals/f4b6b1a3-c261-448d-af63-60341d612f77","signal_json_url":"https://onlylabs.fyi/signals/f4b6b1a3-c261-448d-af63-60341d612f77/signal.json","text":"model_released · arcee-ai/arcee-blitz-caller-beta · signal_desk=releases · occurred_at=2025-02-21T06:53:45+00:00 · url=https://huggingface.co/arcee-ai/arcee-blitz-caller-beta · hf_downloads=4 · hf_likes=1 · hf_params=23572423680 · license=apache-2.0 · raw={\"derived_reason\":\"first-party-finetune\"}"},{"ref":"E37","kind":"event","title":"arcee-ai/trinity-large-tech-report","date":"2026-01-27T20:08:46+00:00","date_source":"source","source_url":"https://github.com/arcee-ai/trinity-large-tech-report","signal_url":"https://onlylabs.fyi/signals/8e750f42-1a87-42cd-a42a-bcb1acb363a5","signal_json_url":"https://onlylabs.fyi/signals/8e750f42-1a87-42cd-a42a-bcb1acb363a5/signal.json","text":"repo_new · arcee-ai/trinity-large-tech-report · signal_desk=repos · occurred_at=2026-01-27T20:08:46+00:00 · url=https://github.com/arcee-ai/trinity-large-tech-report · stars=125 · raw={\"repo\":\"arcee-ai/trinity-large-tech-report\"}"},{"ref":"E38","kind":"event","title":"arcee-ai/token.js","date":"2025-02-11T03:09:39+00:00","date_source":"source","source_url":"https://github.com/arcee-ai/token.js","signal_url":"https://onlylabs.fyi/signals/f3e69798-2c5d-4ffe-b093-6fcccaf8ed63","signal_json_url":"https://onlylabs.fyi/signals/f3e69798-2c5d-4ffe-b093-6fcccaf8ed63/signal.json","text":"repo_forked · arcee-ai/token.js · signal_desk=forks · occurred_at=2025-02-11T03:09:39+00:00 · url=https://github.com/arcee-ai/token.js · stars=3 · raw={\"repo\":\"arcee-ai/token.js\",\"parent\":\"token-js/token.js\"}"},{"ref":"E39","kind":"event","title":"arcee-ai/entropix","date":"2024-10-14T23:02:30+00:00","date_source":"source","source_url":"https://github.com/arcee-ai/entropix","signal_url":"https://onlylabs.fyi/signals/c9ea6498-8bac-4851-97b3-b6b4c5750071","signal_json_url":"https://onlylabs.fyi/signals/c9ea6498-8bac-4851-97b3-b6b4c5750071/signal.json","text":"repo_forked · arcee-ai/entropix · signal_desk=forks · occurred_at=2024-10-14T23:02:30+00:00 · url=https://github.com/arcee-ai/entropix · stars=3 · raw={\"repo\":\"arcee-ai/entropix\",\"parent\":\"xjdr-alt/entropix\"}"},{"ref":"E40","kind":"event","title":"arcee-ai/optillm","date":"2024-10-04T03:04:29+00:00","date_source":"source","source_url":"https://github.com/arcee-ai/optillm","signal_url":"https://onlylabs.fyi/signals/4819afbd-8628-408a-80b8-64b79c40f0f8","signal_json_url":"https://onlylabs.fyi/signals/4819afbd-8628-408a-80b8-64b79c40f0f8/signal.json","text":"repo_forked · arcee-ai/optillm · signal_desk=forks · occurred_at=2024-10-04T03:04:29+00:00 · url=https://github.com/arcee-ai/optillm · stars=3 · raw={\"repo\":\"arcee-ai/optillm\",\"parent\":\"algorithmicsuperintelligence/optillm\"}"},{"ref":"E41","kind":"event","title":"arcee-ai/optillm-upstream","date":"2024-10-24T23:39:59+00:00","date_source":"source","source_url":"https://github.com/arcee-ai/optillm-upstream","signal_url":"https://onlylabs.fyi/signals/11fb9a0b-83cd-473e-881a-936154cdae66","signal_json_url":"https://onlylabs.fyi/signals/11fb9a0b-83cd-473e-881a-936154cdae66/signal.json","text":"repo_forked · arcee-ai/optillm-upstream · signal_desk=forks · occurred_at=2024-10-24T23:39:59+00:00 · url=https://github.com/arcee-ai/optillm-upstream · stars=2 · raw={\"repo\":\"arcee-ai/optillm-upstream\",\"parent\":\"algorithmicsuperintelligence/optillm\"}"},{"ref":"E42","kind":"event","title":"arcee-ai/dify-playground-frontendd","date":"2024-10-24T19:12:17+00:00","date_source":"source","source_url":"https://github.com/arcee-ai/dify-playground-frontendd","signal_url":"https://onlylabs.fyi/signals/994201ed-dbf7-46ec-ab1e-417e4d1bc2dc","signal_json_url":"https://onlylabs.fyi/signals/994201ed-dbf7-46ec-ab1e-417e4d1bc2dc/signal.json","text":"repo_forked · arcee-ai/dify-playground-frontendd · signal_desk=forks · occurred_at=2024-10-24T19:12:17+00:00 · url=https://github.com/arcee-ai/dify-playground-frontendd · stars=2 · raw={\"repo\":\"arcee-ai/dify-playground-frontendd\",\"parent\":\"langgenius/dify\"}"},{"ref":"E43","kind":"event","title":"arcee-ai/Megatron-LM-Llama-70B","date":"2024-05-22T17:40:11+00:00","date_source":"source","source_url":"https://github.com/arcee-ai/Megatron-LM-Llama-70B","signal_url":"https://onlylabs.fyi/signals/ca2a6f89-1577-4bdf-9f9f-b5c3c87b3c19","signal_json_url":"https://onlylabs.fyi/signals/ca2a6f89-1577-4bdf-9f9f-b5c3c87b3c19/signal.json","text":"repo_forked · arcee-ai/Megatron-LM-Llama-70B · signal_desk=forks · occurred_at=2024-05-22T17:40:11+00:00 · url=https://github.com/arcee-ai/Megatron-LM-Llama-70B · stars=2 · raw={\"repo\":\"arcee-ai/Megatron-LM-Llama-70B\",\"parent\":\"NVIDIA/Megatron-LM\"}"},{"ref":"E44","kind":"event","title":"arcee-ai/llm-autoeval","date":"2024-02-28T03:16:00+00:00","date_source":"source","source_url":"https://github.com/arcee-ai/llm-autoeval","signal_url":"https://onlylabs.fyi/signals/ce5c87c6-b574-45a9-bbee-e4d5bfa7c199","signal_json_url":"https://onlylabs.fyi/signals/ce5c87c6-b574-45a9-bbee-e4d5bfa7c199/signal.json","text":"repo_forked · arcee-ai/llm-autoeval · signal_desk=forks · occurred_at=2024-02-28T03:16:00+00:00 · url=https://github.com/arcee-ai/llm-autoeval · stars=2 · raw={\"repo\":\"arcee-ai/llm-autoeval\",\"parent\":\"mlabonne/llm-autoeval\"}"},{"ref":"E45","kind":"event","title":"arcee-ai/pipelines","date":"2025-02-17T16:02:04+00:00","date_source":"source","source_url":"https://github.com/arcee-ai/pipelines","signal_url":"https://onlylabs.fyi/signals/739cdf1f-418d-457c-8c03-f211d133d002","signal_json_url":"https://onlylabs.fyi/signals/739cdf1f-418d-457c-8c03-f211d133d002/signal.json","text":"repo_forked · arcee-ai/pipelines · signal_desk=forks · occurred_at=2025-02-17T16:02:04+00:00 · url=https://github.com/arcee-ai/pipelines · stars=1 · raw={\"repo\":\"arcee-ai/pipelines\",\"parent\":\"open-webui/pipelines\"}"},{"ref":"E46","kind":"event","title":"arcee-ai/open-instruct","date":"2024-12-11T23:54:41+00:00","date_source":"source","source_url":"https://github.com/arcee-ai/open-instruct","signal_url":"https://onlylabs.fyi/signals/9d00ab53-6cd6-4142-8abe-01bfec477f76","signal_json_url":"https://onlylabs.fyi/signals/9d00ab53-6cd6-4142-8abe-01bfec477f76/signal.json","text":"repo_forked · arcee-ai/open-instruct · signal_desk=forks · occurred_at=2024-12-11T23:54:41+00:00 · url=https://github.com/arcee-ai/open-instruct · stars=1 · raw={\"repo\":\"arcee-ai/open-instruct\",\"parent\":\"allenai/open-instruct\"}"},{"ref":"E47","kind":"event","title":"arcee-ai/chat-ui","date":"2024-08-19T14:57:12+00:00","date_source":"source","source_url":"https://github.com/arcee-ai/chat-ui","signal_url":"https://onlylabs.fyi/signals/bcf97310-e009-4f16-befe-2771aadc0bf2","signal_json_url":"https://onlylabs.fyi/signals/bcf97310-e009-4f16-befe-2771aadc0bf2/signal.json","text":"repo_forked · arcee-ai/chat-ui · signal_desk=forks · occurred_at=2024-08-19T14:57:12+00:00 · url=https://github.com/arcee-ai/chat-ui · stars=1 · raw={\"repo\":\"arcee-ai/chat-ui\",\"parent\":\"huggingface/chat-ui\"}"},{"ref":"E48","kind":"event","title":"arcee-ai/Pai-Megatron-Patch-Llama3-70B","date":"2024-05-19T06:01:51+00:00","date_source":"source","source_url":"https://github.com/arcee-ai/Pai-Megatron-Patch-Llama3-70B","signal_url":"https://onlylabs.fyi/signals/7706f2d3-6244-4804-9dc2-35b7b18da0f8","signal_json_url":"https://onlylabs.fyi/signals/7706f2d3-6244-4804-9dc2-35b7b18da0f8/signal.json","text":"repo_forked · arcee-ai/Pai-Megatron-Patch-Llama3-70B · signal_desk=forks · occurred_at=2024-05-19T06:01:51+00:00 · url=https://github.com/arcee-ai/Pai-Megatron-Patch-Llama3-70B · stars=1 · raw={\"repo\":\"arcee-ai/Pai-Megatron-Patch-Llama3-70B\",\"parent\":\"alibaba/Pai-Megatron-Patch\"}"},{"ref":"E49","kind":"event","title":"Distilling Kimi Delta Attention Into Afm 4 5b And The Tool We Used To Do It","date":"2025-12-15T00:00:00.000Z","date_source":"page.visible_date","source_url":"https://www.arcee.ai/blog/distilling-kimi-delta-attention-into-afm-4-5b-and-the-tool-we-used-to-do-it","signal_url":"https://onlylabs.fyi/signals/15aeaa90-183a-4111-ac12-71167ee53dfa","signal_json_url":"https://onlylabs.fyi/signals/15aeaa90-183a-4111-ac12-71167ee53dfa/signal.json","text":"post_published · Distilling Kimi Delta Attention Into Afm 4 5b And The Tool We Used To Do It · signal_desk=talking · occurred_at=2025-12-15T00:00:00.000Z · url=https://www.arcee.ai/blog/distilling-kimi-delta-attention-into-afm-4-5b-and-the-tool-we-used-to-do-it"},{"ref":"E50","kind":"event","title":"arcee-ai/prime-rl","date":"2025-11-11T00:08:20+00:00","date_source":"source","source_url":"https://github.com/arcee-ai/prime-rl","signal_url":"https://onlylabs.fyi/signals/8da333a0-40e4-4208-b8bb-5cdd92073433","signal_json_url":"https://onlylabs.fyi/signals/8da333a0-40e4-4208-b8bb-5cdd92073433/signal.json","text":"repo_forked · arcee-ai/prime-rl · signal_desk=forks · occurred_at=2025-11-11T00:08:20+00:00 · url=https://github.com/arcee-ai/prime-rl · raw={\"repo\":\"arcee-ai/prime-rl\",\"parent\":\"PrimeIntellect-ai/prime-rl\"}"},{"ref":"E51","kind":"event","title":"Extending Afm 4 5b To 64k Context Length","date":"2025-06-23T00:00:00.000Z","date_source":"page.visible_date","source_url":"https://www.arcee.ai/blog/extending-afm-4-5b-to-64k-context-length","signal_url":"https://onlylabs.fyi/signals/10f8de37-74f2-4642-b3ca-e9e9e14942bd","signal_json_url":"https://onlylabs.fyi/signals/10f8de37-74f2-4642-b3ca-e9e9e14942bd/signal.json","text":"post_published · Extending Afm 4 5b To 64k Context Length · signal_desk=talking · occurred_at=2025-06-23T00:00:00.000Z · url=https://www.arcee.ai/blog/extending-afm-4-5b-to-64k-context-length"},{"ref":"E52","kind":"event","title":"How Arcee Ai Helped Madeline Co Build A World Class Reasoning Model From First Principles","date":"2025-06-06T00:00:00.000Z","date_source":"page.visible_date","source_url":"https://www.arcee.ai/blog/how-arcee-ai-helped-madeline-co-build-a-world-class-reasoning-model-from-first-principles","signal_url":"https://onlylabs.fyi/signals/ed741bba-d676-4f60-b0ae-a24f03d7d99c","signal_json_url":"https://onlylabs.fyi/signals/ed741bba-d676-4f60-b0ae-a24f03d7d99c/signal.json","text":"post_published · How Arcee Ai Helped Madeline Co Build A World Class Reasoning Model From First Principles · signal_desk=talking · occurred_at=2025-06-06T00:00:00.000Z · url=https://www.arcee.ai/blog/how-arcee-ai-helped-madeline-co-build-a-world-class-reasoning-model-from-first-principles"},{"ref":"E53","kind":"event","title":"Enriching Inventory Data With Arcee Conductor","date":"2025-05-09T00:00:00.000Z","date_source":"page.visible_date","source_url":"https://www.arcee.ai/blog/enriching-inventory-data-with-arcee-conductor","signal_url":"https://onlylabs.fyi/signals/23a84876-b967-4d7b-bf9d-23e74475c117","signal_json_url":"https://onlylabs.fyi/signals/23a84876-b967-4d7b-bf9d-23e74475c117/signal.json","text":"post_published · Enriching Inventory Data With Arcee Conductor · signal_desk=talking · occurred_at=2025-05-09T00:00:00.000Z · url=https://www.arcee.ai/blog/enriching-inventory-data-with-arcee-conductor"},{"ref":"E54","kind":"event","title":"How Knowledge Distillation Works And When To Use It","date":"2025-02-04T00:00:00.000Z","date_source":"page.visible_date","source_url":"https://www.arcee.ai/blog/how-knowledge-distillation-works-and-when-to-use-it","signal_url":"https://onlylabs.fyi/signals/5d4ee532-cfa8-49d4-b534-ca4d655b8a94","signal_json_url":"https://onlylabs.fyi/signals/5d4ee532-cfa8-49d4-b534-ca4d655b8a94/signal.json","text":"post_published · How Knowledge Distillation Works And When To Use It · signal_desk=talking · occurred_at=2025-02-04T00:00:00.000Z · url=https://www.arcee.ai/blog/how-knowledge-distillation-works-and-when-to-use-it"},{"ref":"E55","kind":"event","title":"Introducing Arcee Vylinh A Powerful 3b Parameter Vietnamese Language Model","date":"2024-11-07T00:00:00.000Z","date_source":"page.visible_date","source_url":"https://www.arcee.ai/blog/introducing-arcee-vylinh-a-powerful-3b-parameter-vietnamese-language-model","signal_url":"https://onlylabs.fyi/signals/aa7b9172-74cc-4df0-b50c-0471a91086a0","signal_json_url":"https://onlylabs.fyi/signals/aa7b9172-74cc-4df0-b50c-0471a91086a0/signal.json","text":"post_published · Introducing Arcee Vylinh A Powerful 3b Parameter Vietnamese Language Model · signal_desk=talking · occurred_at=2024-11-07T00:00:00.000Z · url=https://www.arcee.ai/blog/introducing-arcee-vylinh-a-powerful-3b-parameter-vietnamese-language-model"},{"ref":"E56","kind":"event","title":"Introducing Arcee Supernova Medius A 14b Model That Rivals A 70b 2","date":"2024-10-11T00:00:00.000Z","date_source":"page.visible_date","source_url":"https://www.arcee.ai/blog/introducing-arcee-supernova-medius-a-14b-model-that-rivals-a-70b-2","signal_url":"https://onlylabs.fyi/signals/f7560627-32cf-4d36-a097-5b3e831053f5","signal_json_url":"https://onlylabs.fyi/signals/f7560627-32cf-4d36-a097-5b3e831053f5/signal.json","text":"post_published · Introducing Arcee Supernova Medius A 14b Model That Rivals A 70b 2 · signal_desk=talking · occurred_at=2024-10-11T00:00:00.000Z · url=https://www.arcee.ai/blog/introducing-arcee-supernova-medius-a-14b-model-that-rivals-a-70b-2"},{"ref":"E57","kind":"event","title":"Introducing Arcee Nova","date":"2024-07-18T00:00:00.000Z","date_source":"page.visible_date","source_url":"https://www.arcee.ai/blog/introducing-arcee-nova","signal_url":"https://onlylabs.fyi/signals/e40f2f44-fda1-4dd0-9e96-d4d323a02906","signal_json_url":"https://onlylabs.fyi/signals/e40f2f44-fda1-4dd0-9e96-d4d323a02906/signal.json","text":"post_published · Introducing Arcee Nova · signal_desk=talking · occurred_at=2024-07-18T00:00:00.000Z · url=https://www.arcee.ai/blog/introducing-arcee-nova"},{"ref":"E58","kind":"event","title":"How Do I Prep My Data To Train An Llm 2","date":"2024-07-09T00:00:00.000Z","date_source":"page.visible_date","source_url":"https://www.arcee.ai/blog/how-do-i-prep-my-data-to-train-an-llm-2","signal_url":"https://onlylabs.fyi/signals/e9dcbabc-7329-4c72-8012-3532a62f5c8b","signal_json_url":"https://onlylabs.fyi/signals/e9dcbabc-7329-4c72-8012-3532a62f5c8b/signal.json","text":"post_published · How Do I Prep My Data To Train An Llm 2 · signal_desk=talking · occurred_at=2024-07-09T00:00:00.000Z · url=https://www.arcee.ai/blog/how-do-i-prep-my-data-to-train-an-llm-2"},{"ref":"E59","kind":"event","title":"Case Study Innovating Domain Adaptation Through Continual Pre Training And Model Merging","date":"2024-03-19T00:00:00.000Z","date_source":"page.visible_date","source_url":"https://www.arcee.ai/blog/case-study-innovating-domain-adaptation-through-continual-pre-training-and-model-merging","signal_url":"https://onlylabs.fyi/signals/3d9ca054-0d41-4c3d-91b0-27e95059f5bf","signal_json_url":"https://onlylabs.fyi/signals/3d9ca054-0d41-4c3d-91b0-27e95059f5bf/signal.json","text":"post_published · Case Study Innovating Domain Adaptation Through Continual Pre Training And Model Merging · signal_desk=talking · occurred_at=2024-03-19T00:00:00.000Z · url=https://www.arcee.ai/blog/case-study-innovating-domain-adaptation-through-continual-pre-training-and-model-merging"},{"ref":"E60","kind":"event","title":"How Model Merging Fits Into Slm System","date":"2024-03-05T00:00:00.000Z","date_source":"page.visible_date","source_url":"https://www.arcee.ai/blog/how-model-merging-fits-into-slm-system","signal_url":"https://onlylabs.fyi/signals/09b4c564-ae29-4744-ae73-ca39e3f9e4ca","signal_json_url":"https://onlylabs.fyi/signals/09b4c564-ae29-4744-ae73-ca39e3f9e4ca/signal.json","text":"post_published · How Model Merging Fits Into Slm System · signal_desk=talking · occurred_at=2024-03-05T00:00:00.000Z · url=https://www.arcee.ai/blog/how-model-merging-fits-into-slm-system"}]}