{"schema_version":"onlylabs.public_analysis.v1","url":"https://onlylabs.fyi/analysis/openbmb","json_url":"https://onlylabs.fyi/analysis/openbmb/analysis.json","evidence_json_url":"https://onlylabs.fyi/analysis/openbmb/evidence.json","generated_at":"2026-06-28T02:25:49.419Z","analysis":{"org_slug":"openbmb","url":"https://onlylabs.fyi/analysis/openbmb","json_url":"https://onlylabs.fyi/analysis/openbmb/analysis.json","evidence_json_url":"https://onlylabs.fyi/analysis/openbmb/evidence.json","dossier_url":"https://onlylabs.fyi/labs/openbmb","org":{"slug":"openbmb","name":"OpenBMB (MiniCPM)","category":"neolab","category_label":"Neolab","homepage_url":"https://www.openbmb.cn"},"title":"OpenBMB (MiniCPM) analysis","summary":"OpenBMB is a university-tethered research-to-product organization—with author affiliations spanning Tsinghua University and Northeastern University —pursuing a clear thesis: compact, on-device frontier AI that ships openly and runs locally. Their portfolio spans language models (MiniCPM series, CPM-Bee), vision-language models (MiniCPM-V, VisCPM), speech synthesis (VoxCPM), agent frameworks (AgentCPM, XAgent,…","markdown":"## Thesis\n\nOpenBMB is a university-tethered research-to-product organization—with author affiliations spanning Tsinghua University and Northeastern University [P3](https://github.com/OpenBMB/SHIFT)—pursuing a clear thesis: **compact, on-device frontier AI that ships openly and runs locally.** Their portfolio spans language models (MiniCPM series, CPM-Bee), vision-language models (MiniCPM-V, VisCPM), speech synthesis (VoxCPM), agent frameworks (AgentCPM, XAgent, AgentVerse, ChatDev, ProAgent, IoA), RAG infrastructure (UltraRAG, VisRAG, SHIFT), training tooling (BMTrain, ModelCenter, ForgeTrain, BMCook), inference optimization (BMInf, cpm_kernels), and evaluation benchmarks (UltraEval, InfiniteBench, ToolBench, MA-ProofBench, AceBench, Omni-DuplexEval). Everything ships under permissive Apache 2.0 or MIT licenses. The evidence in this pack covers a period of accelerating productization (May–June 2026): OpenBMB is pushing beyond model releases into consumer-facing desktop and mobile agent applications—PilotDeck, MiniCPM-Desk-Pet, and MiniCPM-V-Apps—while simultaneously shipping new model families (MiniCPM5, BitCPM-CANN, SciCore) and research on RAG knowledge conflicts, theorem proving, and agent benchmarks. The signal is of a lab simultaneously deepening its research pipeline and building end-user distribution channels on mobile and desktop platforms.\n\n## Signal desks\n\n### Hiring\n\nNo cited evidence in this pack. No job postings, career pages, role descriptions, or hiring announcements appear in any of the supplied sources.\n\n### Forks\n\n- **EdgeClaw** — Forked from `openclaw/openclaw`, an open-source agent/claw framework. 1,223 stars on the fork. Indicates active inspection of agent orchestration infrastructure for potential integration or adaptation into OpenBMB's own agent stack (which already includes XAgent, AgentVerse, ChatDev, ProAgent, AgentCPM, and IoA). [E54](https://github.com/OpenBMB/EdgeClaw)\n- **sglang** — Forked from `sgl-project/sglang`, a high-performance LLM inference/serving engine. 14 stars on the fork. Consistent with OpenBMB's stated compatibility: MiniCPM5-1B explicitly advertises SGLang support alongside vLLM and Transformers. The fork suggests internal work on inference serving optimization. [E60](https://github.com/OpenBMB/sglang) [W2](https://decrypt.co/369068/openbmb-minicpm5-half-gigabyte-ai-model-local-agents-phone?amp=1)\n\n### Releases\n\n- **MiniCPM5-1B family** — A 1.08B-parameter dense Llama-class model with 128K context, English+Chinese, hybrid reasoning toggle, native XML tool calling, MCP support [W1](https://theaibench.ai/changes/drops/2026-05-22-minicpm5-1b/) [W2](https://decrypt.co/369068/openbmb-minicpm5-half-gigabyte-ai-model-local-agents-phone?amp=1) [W3](https://www.creativeainews.com/blog/minicpm5-1b-openbmb-sota-1b-local-llm-2026/) [E4](https://huggingface.co/openbmb/MiniCPM5-1B). Accompanied by SFT variant [E14](https://huggingface.co/openbmb/MiniCPM5-1B-SFT) and Base variant [E36](https://huggingface.co/openbmb/MiniCPM5-1B-Base). Released late May 2026 alongside the MiniCPM 5.0 repository milestone [E30](https://github.com/OpenBMB/MiniCPM/releases/tag/5.0). Claims 1B-class SOTA [W1](https://theaibench.ai/changes/drops/2026-05-22-minicpm5-1b/) [W3](https://www.creativeainews.com/blog/minicpm5-1b-openbmb-sota-1b-local-llm-2026/). Apache 2.0, runs on CPU [W3](https://www.creativeainews.com/blog/minicpm5-1b-openbmb-sota-1b-local-llm-2026/).\n- **MiniCPM-V-4.6** — A 1B-parameter vision-language model optimized for native on-device execution on iPhone, Android, and HarmonyOS [W4](https://indicops.com/the-smartphone-just-became-an-ai-computer-openbmbs-minicpm-v-4-6-pushes-vision-ai-fully-on-device/) [E2](https://huggingface.co/openbmb/MiniCPM-V-4.6). Built with the LLaVA-UHD v4 architecture optimization pipeline [W4](https://indicops.com/the-smartphone-just-became-an-ai-computer-openbmbs-minicpm-v-4-6-pushes-vision-ai-fully-on-device/). Ships with fine-tuning recipes, SWIFT integration, and LLaMA-Factory support [W4](https://indicops.com/the-smartphone-just-became-an-ai-computer-openbmbs-minicpm-v-4-6-pushes-vision-ai-fully-on-device/). A \"Thinking\" variant released May 2026 [E15](https://huggingface.co/openbmb/MiniCPM-V-4.6-Thinking).\n- **PilotDeck Desktop** — A task-oriented AI agent productivity platform that reached 3,693 GitHub stars in its first month [E16](https://github.com/OpenBMB/PilotDeck). Two releases in the pack: v0.1.0 (June 10) and v260623 (June 23) [E22](https://github.com/OpenBMB/PilotDeck/releases/tag/v0.1.0) [E17](https://github.com/OpenBMB/PilotDeck/releases/tag/v260623) [P2](https://github.com/OpenBMB/PilotDeck/releases/tag/v260623). The v260623 release added Feishu and WeChat IM channel integration, Cron/Always-on workflows, improved streaming UX, and sub-agent cards [P2](https://github.com/OpenBMB/PilotDeck/releases/tag/v260623). Ships as DMG for macOS ARM64 and EXE for Windows x64 [P2](https://github.com/OpenBMB/PilotDeck/releases/tag/v260623).\n- **MiniCPM-Desk-Pet** — A local-first desktop pet powered by MiniCPM5 [E37](https://github.com/OpenBMB/MiniCPM-Desk-Pet). Rapid release cadence: v0.7.1 → v0.7.2 → v0.7.3 → v0.10.0 within approximately one month [E33](https://github.com/OpenBMB/MiniCPM-Desk-Pet/releases/tag/v0.7.1) [E29](https://github.com/OpenBMB/MiniCPM-Desk-Pet/releases/tag/v0.7.2) [E28](https://github.com/OpenBMB/MiniCPM-Desk-Pet/releases/tag/v0.7.3) [E20](https://github.com/OpenBMB/MiniCPM-Desk-Pet/releases/tag/v0.10.0) [P4](https://github.com/OpenBMB/MiniCPM-Desk-Pet/releases/tag/v0.10.0). v0.10.0 added a Hamster theme, improved post-conversation summarization, inference engine (sidecar) stability, and Windows ARM64 support [P4](https://github.com/OpenBMB/MiniCPM-Desk-Pet/releases/tag/v0.10.0). Cross-platform: macOS (Metal GPU), Windows x64 (CPU/Vulkan), Windows ARM64 (CPU) [P4](https://github.com/OpenBMB/MiniCPM-Desk-Pet/releases/tag/v0.10.0).\n- **MiniCPM-V-Apps** — Mobile application releases for Android and HarmonyOS across multiple versions (v1.7 through v2.3) [E24](https://github.com/OpenBMB/MiniCPM-V-Apps/releases/tag/v2.3) [E31](https://github.com/OpenBMB/MiniCPM-V-Apps/releases/tag/harmonyos-v2.2) [E32](https://github.com/OpenBMB/MiniCPM-V-Apps/releases/tag/android-v2.2) [E40](https://github.com/OpenBMB/MiniCPM-V-Apps/releases/tag/harmonyos-v2.0) [E41](https://github.com/OpenBMB/MiniCPM-V-Apps/releases/tag/android-v2.0) [E42](https://github.com/OpenBMB/MiniCPM-V-Apps/releases/tag/android-v1.9) [E43](https://github.com/OpenBMB/MiniCPM-V-Apps/releases/tag/harmonyos-v1.2) [E44](https://github.com/OpenBMB/MiniCPM-V-Apps/releases/tag/harmonyos-v1.1) [E45](https://github.com/OpenBMB/MiniCPM-V-Apps/releases/tag/android-v1.7). Parallel Android and HarmonyOS track indicates dual-ecosystem strategy.\n- **AgentCPM series** — AgentCPM-Report (8B deep research agent, January 2026) [E11](https://huggingface.co/openbmb/AgentCPM-Report) [P1](https://huggingface.co/openbmb/AgentCPM-Report/raw/main/README.md) and AgentCPM-Explore (4B long-horizon agent, January 2026) [E8](https://huggingface.co/openbmb/AgentCPM-Explore). AgentCPM-GUI, an on-device GUI agent for Android apps (May 2025) [E53](https://github.com/OpenBMB/AgentCPM-GUI). The Report model uses a \"Writing As Reasoning Policy\" (WARP) to alternate between evidence-based drafting and reasoning-driven deepening [P1](https://huggingface.co/openbmb/AgentCPM-Report/raw/main/README.md).\n- **Speech models** — VoxCPM2 (2.3B params, Apache 2.0, 584K downloads, 1,434 likes) [E1](https://huggingface.co/openbmb/VoxCPM2), VoxCPM1.5 [E10](https://huggingface.co/openbmb/VoxCPM1.5), VoxCPM-0.5B [E5](https://huggingface.co/openbmb/VoxCPM-0.5B). VoxCPM repository: 31,827 stars [E12](https://github.com/OpenBMB/VoxCPM). VoxCPM 2.0.3 release in May 2026 [E46](https://github.com/OpenBMB/VoxCPM/releases/tag/2.0.3).\n- **BitCPM-CANN series** — Three quantized/compressed models at 0.5B, 1B, 3B, and 8B scales, all released May 2026 [E26](https://huggingface.co/openbmb/BitCPM-CANN-0.5B) [E27](https://huggingface.co/openbmb/BitCPM-CANN-1B) [E25](https://huggingface.co/openbmb/BitCPM-CANN-3B) [E13](https://huggingface.co/openbmb/BitCPM-CANN-8B). Signals a model compression and edge-deployment product line.\n- **SciCore** — Domain-specific science models: SciCore-Omics (omics/molecular biology) [E18](https://huggingface.co/openbmb/SciCore-Omics) and SciCore-Mol (chemistry) [E47](https://huggingface.co/openbmb/SciCore-Mol).\n- **Infrastructure releases** — UltraRAG v0.3.0.2 (April 2026) [E57](https://github.com/OpenBMB/UltraRAG/releases/tag/v0.3.0.2); ArcLight v1.1 (May 2026) [E35](https://github.com/OpenBMB/ArcLight/releases/tag/v1.1); VoxCPM 2.0.3 (May 2026) [E46](https://github.com/OpenBMB/VoxCPM/releases/tag/2.0.3).\n- **Legacy reference repos persist**: BMTrain (624 stars, last pushed April 2026) [P8](https://github.com/OpenBMB/BMTrain), BMInf (585 stars), BMCook (169 stars), ModelCenter (271 stars), CPM-Bee (2,406 stars), BMTools (2,773 stars), ToolBench (5,663 stars) [P6](https://github.com/OpenBMB/BMInf) [P13](https://github.com/OpenBMB/BMCook) [P12](https://github.com/OpenBMB/ModelCenter) [P16](https://github.com/OpenBMB/CPM-Bee) [P15](https://github.com/OpenBMB/BMTools) [P20](https://github.com/OpenBMB/ToolBench). These are not actively maintained but signal sustained community interest.\n\n### Talking\n\n- **MiniCPM5-1B launch coverage** — The AI Bench characterized it as a \"three-stage SFT → RL → On-Policy Distillation pipeline\" landing at 1B scale, claiming SOTA against LFM2.5-1.2B-Thinking, Qwen3-0.6B/think, and Qwen3.5-0.8B/think [W1](https://theaibench.ai/changes/drops/2026-05-22-minicpm5-1b/). Creative AI News emphasized the model beating the 2B-scale Qwen3.5-2B on small-model benchmarks and highlighted 131K token context and CPU-only operation [W3](https://www.creativeainews.com/blog/minicpm5-1b-openbmb-sota-1b-local-llm-2026/). Decrypt focused on the on-device agent story: fitting on smartphone memory, native tool calling, MCP support, Apache 2.0 licensing, and vLLM/SGLang compatibility [W2](https://decrypt.co/369068/openbmb-minicpm5-half-gigabyte-ai-model-local-agents-phone?amp=1).\n- **MiniCPM-V 4.6 framing** — Indicops positioned the release as \"the smartphone just became an AI computer,\" emphasizing fully on-device vision AI, the LLaVA-UHD v4 architecture, triple-platform deployment code (iPhone, Android, HarmonyOS), and fine-tuning cookbooks [W4](https://indicops.com/the-smartphone-just-became-an-ai-computer-openbmbs-minicpm-v-4-6-pushes-vision-ai-fully-on-device/). The narrative is explicitly anti-cloud: \"Unlike most multimodal AI systems that depend heavily on cloud infrastructure.\"\n- **OpenBMB's own GitHub messaging** — The MiniCPM repo frames MiniCPM5-1B as designed for \"local assistants, coding agents, tool-use workflows, and reasoning where a compact model is preferred\" with both Think/No Think chat modes [W5](https://github.com/OpenBMB/MiniCPM). The AgentCPM-Report model card claims \"Gemini-2.5-pro-DeepResearch Level Local DeepResearch,\" positioning against Google's cloud-only offering [P1](https://huggingface.co/openbmb/AgentCPM-Report/raw/main/README.md).\n- **No HN traction of note** in this pack. The VoxCPM repo garnered only 3 points/0 comments on HN [E12](https://github.com/OpenBMB/VoxCPM); IoA got 3 points/1 comment [E48](https://github.com/OpenBMB/IoA). Community attention appears concentrated on GitHub stars and Hugging Face downloads rather than Hacker News discussion.\n\n## Shipping\n\nOpenBMB's shipping velocity is intense in the April–June 2026 window, with five discernible product tracks:\n\n1. **On-device language models**: MiniCPM5-1B (and base/SFT variants) [E4](https://huggingface.co/openbmb/MiniCPM5-1B) [E14](https://huggingface.co/openbmb/MiniCPM5-1B-SFT) [E36](https://huggingface.co/openbmb/MiniCPM5-1B-Base) as the flagship compact LLM; BitCPM-CANN at four sizes (0.5B–8B) as the compressed/quantized product line [E26](https://huggingface.co/openbmb/BitCPM-CANN-0.5B) [E27](https://huggingface.co/openbmb/BitCPM-CANN-1B) [E25](https://huggingface.co/openbmb/BitCPM-CANN-3B) [E13](https://huggingface.co/openbmb/BitCPM-CANN-8B); MiniCPM4.1-8B and Eagle3 variant as prior-generation holdovers [E9](https://huggingface.co/openbmb/MiniCPM4.1-8B) [E58](https://huggingface.co/openbmb/MiniCPM4.1-8B-Eagle3).\n\n2. **On-device vision-language models**: MiniCPM-V-4.6 and MiniCPM-V-4.6-Thinking [E2](https://huggingface.co/openbmb/MiniCPM-V-4.6) [E15](https://huggingface.co/openbmb/MiniCPM-V-4.6-Thinking) with full deployment code for three mobile OS platforms [W4](https://indicops.com/the-smartphone-just-became-an-ai-computer-openbmbs-minicpm-v-4-6-pushes-vision-ai-fully-on-device/). Preceded by MiniCPM-V-4.5 (8.7B) [E3](https://huggingface.co/openbmb/MiniCPM-V-4_5) and MiniCPM-V-4 (4.1B) [E7](https://huggingface.co/openbmb/MiniCPM-V-4), showing consistent range compression toward smaller, device-friendly footprints.\n\n3. **Speech synthesis**: VoxCPM2 (2.3B params, 584K downloads, 1,434 likes — the highest-traction model in the pack) [E1](https://huggingface.co/openbmb/VoxCPM2), plus smaller VoxCPM1.5 and VoxCPM-0.5B [E10](https://huggingface.co/openbmb/VoxCPM1.5) [E5](https://huggingface.co/openbmb/VoxCPM-0.5B).\n\n4. **Desktop + mobile agent platforms**: PilotDeck Desktop (macOS + Windows) [E16](https://github.com/OpenBMB/PilotDeck) [E17](https://github.com/OpenBMB/PilotDeck/releases/tag/v260623) [P2](https://github.com/OpenBMB/PilotDeck/releases/tag/v260623), MiniCPM-Desk-Pet (cross-platform desktop companion) [E37](https://github.com/OpenBMB/MiniCPM-Desk-Pet) [P4](https://github.com/OpenBMB/MiniCPM-Desk-Pet/releases/tag/v0.10.0), MiniCPM-V-Apps (Android + HarmonyOS) [E24](https://github.com/OpenBMB/MiniCPM-V-Apps/releases/tag/v2.3) [E31](https://github.com/OpenBMB/MiniCPM-V-Apps/releases/tag/harmonyos-v2.2) [E32](https://github.com/OpenBMB/MiniCPM-V-Apps/releases/tag/android-v2.2), AgentCPM-GUI (Android GUI agent) [E53](https://github.com/OpenBMB/AgentCPM-GUI). PilotDeck in particular signals serious product investment: IM channel integration (Feishu, WeChat), Cron workflows, and sub-agent orchestration [P2](https://github.com/OpenBMB/PilotDeck/releases/tag/v260623).\n\n5. **Agent models and frameworks**: AgentCPM-Report (8B deep research) [E11](https://huggingface.co/openbmb/AgentCPM-Report) [P1](https://huggingface.co/openbmb/AgentCPM-Report/raw/main/README.md), AgentCPM-Explore (4B agent benchmark model) [E8](https://huggingface.co/openbmb/AgentCPM-Explore), UltraRAG framework (5,610 stars) [E34](https://github.com/OpenBMB/UltraRAG) [E57](https://github.com/OpenBMB/UltraRAG/releases/tag/v0.3.0.2), and the legacy agent ecosystem (XAgent, ChatDev, AgentVerse, ProAgent, IoA) [P25](https://github.com/OpenBMB/XAgent) [P23](https://github.com/OpenBMB/ChatDev) [P18](https://github.com/OpenBMB/AgentVerse) [P26](https://github.com/OpenBMB/ProAgent) [E48](https://github.com/OpenBMB/IoA).\n\n6. **Benchmarks and evaluation**: MA-ProofBench (Lean 4, 200 formal analysis problems) [P5](https://github.com/OpenBMB/MA-ProofBench) [E21](https://github.com/OpenBMB/MA-ProofBench), AceBench [E23](https://github.com/OpenBMB/AceBench), Omni-DuplexEval [E39](https://github.com/OpenBMB/Omni-DuplexEval), plus maintained evaluation infrastructure (UltraEval, InfiniteBench, ToolBench) [P27](https://github.com/OpenBMB/UltraEval) [P28](https://github.com/OpenBMB/InfiniteBench) [P20](https://github.com/OpenBMB/ToolBench).\n\n7. **Research repos**: SHIFT (RAG knowledge conflict mitigation via gate-modulated activation steering, 3 stars, June 2026) [P3](https://github.com/OpenBMB/SHIFT) [E19](https://github.com/OpenBMB/SHIFT); ForgeTrain (training infrastructure, 239 stars, May 2026) [E38](https://github.com/OpenBMB/ForgeTrain); plus frozen legacy infrastructure (BMTrain, BMInf, BMCook, ModelCenter) [P8](https://github.com/OpenBMB/BMTrain) [P6](https://github.com/OpenBMB/BMInf) [P13](https://github.com/OpenBMB/BMCook) [P12](https://github.com/OpenBMB/ModelCenter).\n\nThe shipping pattern reveals a lab that simultaneously maintains: (a) a consumer product layer (PilotDeck, Desk-Pet, V-Apps), (b) a compact model factory (MiniCPM5, BitCPM-CANN, MiniCPM-V), (c) a speech pipeline (VoxCPM), (d) an agent research program (AgentCPM, SHIFT, benchmarks), and (e) domain-specific models (SciCore). This breadth with a small-parameter focus is unusual among frontier labs and suggests a deliberate strategy of capturing on-device and edge deployment markets rather than competing at the largest scales.\n\n## Research themes\n\nThree research themes are salient in this evidence pack:\n\n1. **On-device and small-model performance** — The MiniCPM5-1B launch centers on matching or exceeding larger models (Qwen3.5-2B) at 1B scale through a three-stage pipeline: SFT → RL → On-Policy Distillation [W1](https://theaibench.ai/changes/drops/2026-05-22-minicpm5-1b/) [W3](https://www.creativeainews.com/blog/minicpm5-1b-openbmb-sota-1b-local-llm-2026/). The BitCPM-CANN series at four sizes (0.5B–8B) suggests active work on quantization-aware training or post-training compression for edge deployment [E26](https://huggingface.co/openbmb/BitCPM-CANN-0.5B) [E27](https://huggingface.co/openbmb/BitCPM-CANN-1B) [E25](https://huggingface.co/openbmb/BitCPM-CANN-3B) [E13](https://huggingface.co/openbmb/BitCPM-CANN-8B). MiniCPM-V-4.6's LLaVA-UHD v4 pipeline is explicitly described as an \"architecture optimization pipeline\" for on-device VLMs [W4](https://indicops.com/the-smartphone-just-became-an-ai-computer-openbmbs-minicpm-v-4-6-pushes-vision-ai-fully-on-device/). This is not just model release; it is a systematic research program in making small models competitive.\n\n2. **Agent autonomy and tool use** — AgentCPM-Report introduces WARP (Writing As Reasoning Policy), an interleaving drafting-and-deepening approach for open-ended research report generation [P1](https://huggingface.co/openbmb/AgentCPM-Report/raw/main/README.md). The claim of \"Gemini-2.5-pro-DeepResearch Level\" performance for a local 8B model [P1](https://huggingface.co/openbmb/AgentCPM-Report/raw/main/README.md) indicates research ambition in closing the agent capability gap between cloud and local. Supporting infrastructure: UltraRAG (5,610 stars) is described as a \"Low-Code MCP Framework for Building Complex and Innovative RAG Pipelines\" [E34](https://github.com/OpenBMB/UltraRAG), and AgentCPM-GUI targets on-device Android GUI automation [E53](https://github.com/OpenBMB/AgentCPM-GUI).\n\n3. **RAG knowledge conflicts** — SHIFT (June 2026) proposes gate-modulated activation steering, adding <0.01% trainable parameters to frozen LLMs to balance retrieved context against parametric knowledge [P3](https://github.com/OpenBMB/SHIFT) [E19](https://github.com/OpenBMB/SHIFT). This connects to the broader RAG theme: VisRAG (parsing-free RAG using VLMs, 968 stars) [E59](https://github.com/OpenBMB/VisRAG) and EVisRAG-7B [E56](https://huggingface.co/openbmb/EVisRAG-7B) suggest sustained investment in retrieval-augmented approaches.\n\n4. **Formal mathematics and benchmarking** — MA-ProofBench introduces the first formal benchmark for theorem proving in mathematical analysis in Lean 4, with 200 problems across undergraduate and PhD tiers [P5](https://github.com/OpenBMB/MA-ProofBench) [E21](https://github.com/OpenBMB/MA-ProofBench). AceBench [E23](https://github.com/OpenBMB/AceBench) and Omni-DuplexEval [E39](https://github.com/OpenBMB/Omni-DuplexEval) round out a new wave of evaluation infrastructure. This continues OpenBMB's established pattern of releasing benchmarks alongside models (UltraEval [P27](https://github.com/OpenBMB/UltraEval), InfiniteBench [P28](https://github.com/OpenBMB/InfiniteBench), ToolBench [P20](https://github.com/OpenBMB/ToolBench)).\n\n5. **Speech synthesis** — VoxCPM2 as a \"Tokenizer-Free TTS for Multilingual Speech Generation, Creative Voice Design, and True-to-Life Cloning\" [E12](https://github.com/OpenBMB/VoxCPM) with 584K Hugging Face downloads [E1](https://huggingface.co/openbmb/VoxCPM2) represents a significant but under-explained research track in the evidence pack.\n\n## Hiring & scaling\n\n**No cited evidence in this pack.** There are no job postings, career pages, team expansion announcements, or hiring signals across any of the supplied sources. This is a notable gap: for an organization simultaneously shipping consumer desktop applications (PilotDeck, Desk-Pet), mobile apps (MiniCPM-V-Apps), model families across text/vision/speech modalities, and multiple research repositories, the absence of hiring evidence suggests either that (a) hiring happens through university channels (Tsinghua/Northeastern) not captured in public job boards, (b) the team is small and not scaling headcount commensurate with output, or (c) hiring signals exist but were not captured in this evidence pack. The P3 author list (8 authors across 2 universities) [P3](https://github.com/OpenBMB/SHIFT) and the MA-ProofBench author pattern [P5](https://github.com/OpenBMB/MA-ProofBench) support hypothesis (a): OpenBMB appears to scale through academic lab structure rather than commercial hiring pipelines.\n\n## Category implications\n\n**Product strategy**: OpenBMB is building an end-user distribution moat through desktop and mobile applications (PilotDeck, MiniCPM-Desk-Pet, MiniCPM-V-Apps) that are model-agnostic containers designed to be powered by their own compact models [P2](https://github.com/OpenBMB/PilotDeck/releases/tag/v260623) [P4](https://github.com/OpenBMB/MiniCPM-Desk-Pet/releases/tag/v0.10.0) [E24](https://github.com/OpenBMB/MiniCPM-V-Apps/releases/tag/v2.3). PilotDeck's Feishu and WeChat IM integration [P2](https://github.com/OpenBMB/PilotDeck/releases/tag/v260623) indicates a deliberate China-ecosystem GTM strategy. MiniCPM-V-Apps' parallel Android and HarmonyOS release tracks [E31](https://github.com/OpenBMB/MiniCPM-V-Apps/releases/tag/harmonyos-v2.2) [E32](https://github.com/OpenBMB/MiniCPM-V-Apps/releases/tag/android-v2.2) [E40](https://github.com/OpenBMB/MiniCPM-V-Apps/releases/tag/harmonyos-v2.0) [E41](https://github.com/OpenBMB/MiniCPM-V-Apps/releases/tag/android-v2.0) signal a bet on Huawei's HarmonyOS ecosystem alongside Google's Android.\n\n**Infrastructure implications**: The sglang fork [E60](https://github.com/OpenBMB/sglang) and advertised vLLM/SGLang/Transformers compatibility for MiniCPM5-1B [W2](https://decrypt.co/369068/openbmb-minicpm5-half-gigabyte-ai-model-local-agents-phone?amp=1) indicate that OpenBMB's model serving strategy relies on community-standard inference frameworks rather than proprietary serving infrastructure. However, BMTrain (624 stars, last pushed April 2026) [P8](https://github.com/OpenBMB/BMTrain) and ForgeTrain (239 stars, May 2026) [E38](https://github.com/OpenBMB/ForgeTrain) show continued investment in custom training infrastructure. The pattern: leverage community inference tooling, but build custom training pipelines.\n\n**Research implications**: MA-ProofBench in Lean 4 [P5](https://github.com/OpenBMB/MA-ProofBench), SHIFT on RAG knowledge conflicts [P3](https://github.com/OpenBMB/SHIFT), and the AgentCPM-Report WARP policy [P1](https://huggingface.co/openbmb/AgentCPM-Report/raw/main/README.md) collectively indicate that OpenBMB maintains active research programs even as productization accelerates. The presence of both ICLR spotlight papers (ToolBench [P20](https://github.com/OpenBMB/ToolBench), VisCPM [P21](https://github.com/OpenBMB/VisCPM)) and ACL publications (UltraEval [P27](https://github.com/OpenBMB/UltraEval), DecT [P17](https://github.com/OpenBMB/DecT)) in the legacy portfolio signals a lab that routes research through top-tier publication venues.\n\n**GTM implications**: OpenBMB's licensing strategy (Apache 2.0 for models, MIT for research repos) is consistent across all recent releases [E4](https://huggingface.co/openbmb/MiniCPM5-1B) [E2](https://huggingface.co/openbmb/MiniCPM-V-4.6) [E1](https://huggingface.co/openbmb/VoxCPM2) [P3](https://github.com/OpenBMB/SHIFT). The General-Model-License repo (8 stars) [P11](https://github.com/OpenBMB/General-Model-License) suggests an earlier exploration of custom licensing that appears to have been abandoned in favor of standard permissive licenses. The \"open-source everything\" posture serves as both a community-growth lever and a competitive differentiator against closed-source or restrictive-license alternatives.\n\n**Talent/team implications**: Evidence is thin. The author lists on SHIFT [P3](https://github.com/OpenBMB/SHIFT) and MA-ProofBench [P5](https://github.com/OpenBMB/MA-ProofBench) show a pipeline from Tsinghua and Northeastern University. No commercial hires, industry veterans, or international expansion signals appear in this pack. The organizational model appears to be an academic research lab with a product-engineering arm, rather than a traditional startup or corporate R&D division.\n\n## Traction highlights\n\n- **ChatDev**: 33,364 GitHub stars, 4,158 forks — the highest-traction OpenBMB project [P23](https://github.com/OpenBMB/ChatDev). Evolved to ChatDev 2.0 (DevAll), a \"Zero-Code Multi-Agent Platform for Developing Everything\" [P23](https://github.com/OpenBMB/ChatDev).\n- **VoxCPM**: 31,827 GitHub stars [E12](https://github.com/OpenBMB/VoxCPM). The VoxCPM2 model: 584,786 Hugging Face downloads, 1,434 likes [E1](https://huggingface.co/openbmb/VoxCPM2) — the highest-traction model release in the pack.\n- **XAgent**: 8,529 stars, 904 forks [P25](https://github.com/OpenBMB/XAgent).\n- **ToolBench**: 5,663 stars, 485 forks; ICLR 2024 spotlight [P20](https://github.com/OpenBMB/ToolBench).\n- **UltraRAG**: 5,610 stars [E34](https://github.com/OpenBMB/UltraRAG).\n- **AgentVerse**: 5,052 stars, 514 forks [P18](https://github.com/OpenBMB/AgentVerse).\n- **PilotDeck**: 3,693 stars within one month of launch [E16](https://github.com/OpenBMB/PilotDeck).\n- **BMTools**: 2,773 stars, 248 forks [P15](https://github.com/OpenBMB/BMTools).\n- **CPM-Bee**: 2,406 stars, 179 forks [P16](https://github.com/OpenBMB/CPM-Bee).\n- **AgentCPM-GUI**: 1,382 stars [E53](https://github.com/OpenBMB/AgentCPM-GUI).\n- **VisCPM**: 1,068 stars, 89 forks; ICLR 2024 spotlight [P21](https://github.com/OpenBMB/VisCPM).\n- **MiniCPM-V-4.6**: 802,002 Hugging Face downloads, 1,127 likes [E2](https://huggingface.co/openbmb/MiniCPM-V-4.6).\n- **MiniCPM5-1B**: 321,584 Hugging Face downloads, 824 likes within days of release [E4](https://huggingface.co/openbmb/MiniCPM5-1B).\n- **EdgeClaw (fork)**: 1,223 stars [E54](https://github.com/OpenBMB/EdgeClaw).\n\nStar counts cluster in the thousands-to-tens-of-thousands range for flagship projects, with Hugging Face download counts in the hundreds of thousands for leading models. Community traction is concentrated in the agent tooling and speech categories, with on-device models (MiniCPM5, MiniCPM-V) showing rapid early adoption.\n\n## Sources\n\nCited page evidence: P1, P2, P3, P4, P5, P6, P8, P11, P12, P13, P15, P16, P17, P18, P20, P21, P23, P25, P26, P27, P28\n\nCited event evidence: E1, E2, E3, E4, E5, E7, E8, E9, E10, E11, E12, E13, E14, E15, E16, E17, E18, E19, E20, E21, E22, E23, E24, E25, E26, E27, E28, E29, E30, E31, E32, E33, E34, E35, E36, E37, E38, E39, E40, E41, E42, E43, E44, E45, E46, E47, E48, E49, E50, E51, E52, E53, E54, E55, E56, E57, E58, E59, E60\n\nCited web evidence: W1, W2, W3, W4, W5","generated_at":"2026-06-27T22:19:26.797+00:00","citations":[{"url":"https://github.com/OpenBMB/SHIFT","path":null,"label":"OpenBMB/SHIFT","type":"external"},{"url":"https://github.com/OpenBMB/EdgeClaw","path":null,"label":"OpenBMB/EdgeClaw","type":"external"},{"url":"https://github.com/OpenBMB/sglang","path":null,"label":"OpenBMB/sglang","type":"external"},{"url":"https://decrypt.co/369068/openbmb-minicpm5-half-gigabyte-ai-model-local-agents-phone?amp=1","path":null,"label":"decrypt.co/369068","type":"external"},{"url":"https://theaibench.ai/changes/drops/2026-05-22-minicpm5-1b/","path":null,"label":"theaibench.ai/changes","type":"external"},{"url":"https://www.creativeainews.com/blog/minicpm5-1b-openbmb-sota-1b-local-llm-2026/","path":null,"label":"creativeainews.com/blog","type":"external"},{"url":"https://huggingface.co/openbmb/MiniCPM5-1B","path":null,"label":"openbmb/MiniCPM5-1B","type":"external"},{"url":"https://huggingface.co/openbmb/MiniCPM5-1B-SFT","path":null,"label":"openbmb/MiniCPM5-1B-SFT","type":"external"},{"url":"https://huggingface.co/openbmb/MiniCPM5-1B-Base","path":null,"label":"openbmb/MiniCPM5-1B-Base","type":"external"},{"url":"https://github.com/OpenBMB/MiniCPM/releases/tag/5.0","path":null,"label":"OpenBMB/MiniCPM","type":"external"},{"url":"https://indicops.com/the-smartphone-just-became-an-ai-computer-openbmbs-minicpm-v-4-6-pushes-vision-ai-fully-on-device/","path":null,"label":"indicops.com/the-smartphone-just-became-an-ai-computer-openbmbs-minicpm-v-4-6-pushes-vision-ai-fully-on-device","type":"external"},{"url":"https://huggingface.co/openbmb/MiniCPM-V-4.6","path":null,"label":"openbmb/MiniCPM-V-4.6","type":"external"},{"url":"https://huggingface.co/openbmb/MiniCPM-V-4.6-Thinking","path":null,"label":"openbmb/MiniCPM-V-4.6-Thinking","type":"external"},{"url":"https://github.com/OpenBMB/PilotDeck","path":null,"label":"OpenBMB/PilotDeck","type":"external"},{"url":"https://github.com/OpenBMB/PilotDeck/releases/tag/v0.1.0","path":null,"label":"OpenBMB/PilotDeck","type":"external"},{"url":"https://github.com/OpenBMB/PilotDeck/releases/tag/v260623","path":null,"label":"OpenBMB/PilotDeck","type":"external"},{"url":"https://github.com/OpenBMB/PilotDeck/releases/tag/v260623","path":null,"label":"OpenBMB/PilotDeck","type":"external"},{"url":"https://github.com/OpenBMB/MiniCPM-Desk-Pet","path":null,"label":"OpenBMB/MiniCPM-Desk-Pet","type":"external"},{"url":"https://github.com/OpenBMB/MiniCPM-Desk-Pet/releases/tag/v0.7.1","path":null,"label":"OpenBMB/MiniCPM-Desk-Pet","type":"external"},{"url":"https://github.com/OpenBMB/MiniCPM-Desk-Pet/releases/tag/v0.7.2","path":null,"label":"OpenBMB/MiniCPM-Desk-Pet","type":"external"},{"url":"https://github.com/OpenBMB/MiniCPM-Desk-Pet/releases/tag/v0.7.3","path":null,"label":"OpenBMB/MiniCPM-Desk-Pet","type":"external"},{"url":"https://github.com/OpenBMB/MiniCPM-Desk-Pet/releases/tag/v0.10.0","path":null,"label":"OpenBMB/MiniCPM-Desk-Pet","type":"external"},{"url":"https://github.com/OpenBMB/MiniCPM-Desk-Pet/releases/tag/v0.10.0","path":null,"label":"OpenBMB/MiniCPM-Desk-Pet","type":"external"},{"url":"https://github.com/OpenBMB/MiniCPM-V-Apps/releases/tag/v2.3","path":null,"label":"OpenBMB/MiniCPM-V-Apps","type":"external"},{"url":"https://github.com/OpenBMB/MiniCPM-V-Apps/releases/tag/harmonyos-v2.2","path":null,"label":"OpenBMB/MiniCPM-V-Apps","type":"external"},{"url":"https://github.com/OpenBMB/MiniCPM-V-Apps/releases/tag/android-v2.2","path":null,"label":"OpenBMB/MiniCPM-V-Apps","type":"external"},{"url":"https://github.com/OpenBMB/MiniCPM-V-Apps/releases/tag/harmonyos-v2.0","path":null,"label":"OpenBMB/MiniCPM-V-Apps","type":"external"},{"url":"https://github.com/OpenBMB/MiniCPM-V-Apps/releases/tag/android-v2.0","path":null,"label":"OpenBMB/MiniCPM-V-Apps","type":"external"},{"url":"https://github.com/OpenBMB/MiniCPM-V-Apps/releases/tag/android-v1.9","path":null,"label":"OpenBMB/MiniCPM-V-Apps","type":"external"},{"url":"https://github.com/OpenBMB/MiniCPM-V-Apps/releases/tag/harmonyos-v1.2","path":null,"label":"OpenBMB/MiniCPM-V-Apps","type":"external"},{"url":"https://github.com/OpenBMB/MiniCPM-V-Apps/releases/tag/harmonyos-v1.1","path":null,"label":"OpenBMB/MiniCPM-V-Apps","type":"external"},{"url":"https://github.com/OpenBMB/MiniCPM-V-Apps/releases/tag/android-v1.7","path":null,"label":"OpenBMB/MiniCPM-V-Apps","type":"external"},{"url":"https://huggingface.co/openbmb/AgentCPM-Report","path":null,"label":"openbmb/AgentCPM-Report","type":"external"},{"url":"https://huggingface.co/openbmb/AgentCPM-Report/raw/main/README.md","path":null,"label":"openbmb/AgentCPM-Report","type":"external"},{"url":"https://huggingface.co/openbmb/AgentCPM-Explore","path":null,"label":"openbmb/AgentCPM-Explore","type":"external"},{"url":"https://github.com/OpenBMB/AgentCPM-GUI","path":null,"label":"OpenBMB/AgentCPM-GUI","type":"external"},{"url":"https://huggingface.co/openbmb/VoxCPM2","path":null,"label":"openbmb/VoxCPM2","type":"external"},{"url":"https://huggingface.co/openbmb/VoxCPM1.5","path":null,"label":"openbmb/VoxCPM1.5","type":"external"},{"url":"https://huggingface.co/openbmb/VoxCPM-0.5B","path":null,"label":"openbmb/VoxCPM-0.5B","type":"external"},{"url":"https://github.com/OpenBMB/VoxCPM","path":null,"label":"OpenBMB/VoxCPM","type":"external"},{"url":"https://github.com/OpenBMB/VoxCPM/releases/tag/2.0.3","path":null,"label":"OpenBMB/VoxCPM","type":"external"},{"url":"https://huggingface.co/openbmb/BitCPM-CANN-0.5B","path":null,"label":"openbmb/BitCPM-CANN-0.5B","type":"external"},{"url":"https://huggingface.co/openbmb/BitCPM-CANN-1B","path":null,"label":"openbmb/BitCPM-CANN-1B","type":"external"},{"url":"https://huggingface.co/openbmb/BitCPM-CANN-3B","path":null,"label":"openbmb/BitCPM-CANN-3B","type":"external"},{"url":"https://huggingface.co/openbmb/BitCPM-CANN-8B","path":null,"label":"openbmb/BitCPM-CANN-8B","type":"external"},{"url":"https://huggingface.co/openbmb/SciCore-Omics","path":null,"label":"openbmb/SciCore-Omics","type":"external"},{"url":"https://huggingface.co/openbmb/SciCore-Mol","path":null,"label":"openbmb/SciCore-Mol","type":"external"},{"url":"https://github.com/OpenBMB/UltraRAG/releases/tag/v0.3.0.2","path":null,"label":"OpenBMB/UltraRAG","type":"external"},{"url":"https://github.com/OpenBMB/ArcLight/releases/tag/v1.1","path":null,"label":"OpenBMB/ArcLight","type":"external"},{"url":"https://github.com/OpenBMB/BMTrain","path":null,"label":"OpenBMB/BMTrain","type":"external"},{"url":"https://github.com/OpenBMB/BMInf","path":null,"label":"OpenBMB/BMInf","type":"external"},{"url":"https://github.com/OpenBMB/BMCook","path":null,"label":"OpenBMB/BMCook","type":"external"},{"url":"https://github.com/OpenBMB/ModelCenter","path":null,"label":"OpenBMB/ModelCenter","type":"external"},{"url":"https://github.com/OpenBMB/CPM-Bee","path":null,"label":"OpenBMB/CPM-Bee","type":"external"},{"url":"https://github.com/OpenBMB/BMTools","path":null,"label":"OpenBMB/BMTools","type":"external"},{"url":"https://github.com/OpenBMB/ToolBench","path":null,"label":"OpenBMB/ToolBench","type":"external"},{"url":"https://github.com/OpenBMB/MiniCPM","path":null,"label":"OpenBMB/MiniCPM","type":"external"},{"url":"https://github.com/OpenBMB/IoA","path":null,"label":"OpenBMB/IoA","type":"external"},{"url":"https://huggingface.co/openbmb/MiniCPM4.1-8B","path":null,"label":"openbmb/MiniCPM4.1-8B","type":"external"},{"url":"https://huggingface.co/openbmb/MiniCPM4.1-8B-Eagle3","path":null,"label":"openbmb/MiniCPM4.1-8B-Eagle3","type":"external"},{"url":"https://huggingface.co/openbmb/MiniCPM-V-4_5","path":null,"label":"openbmb/MiniCPM-V-4_5","type":"external"},{"url":"https://huggingface.co/openbmb/MiniCPM-V-4","path":null,"label":"openbmb/MiniCPM-V-4","type":"external"},{"url":"https://github.com/OpenBMB/UltraRAG","path":null,"label":"OpenBMB/UltraRAG","type":"external"},{"url":"https://github.com/OpenBMB/XAgent","path":null,"label":"OpenBMB/XAgent","type":"external"},{"url":"https://github.com/OpenBMB/ChatDev","path":null,"label":"OpenBMB/ChatDev","type":"external"},{"url":"https://github.com/OpenBMB/AgentVerse","path":null,"label":"OpenBMB/AgentVerse","type":"external"},{"url":"https://github.com/OpenBMB/ProAgent","path":null,"label":"OpenBMB/ProAgent","type":"external"},{"url":"https://github.com/OpenBMB/MA-ProofBench","path":null,"label":"OpenBMB/MA-ProofBench","type":"external"},{"url":"https://github.com/OpenBMB/MA-ProofBench","path":null,"label":"OpenBMB/MA-ProofBench","type":"external"},{"url":"https://github.com/OpenBMB/AceBench","path":null,"label":"OpenBMB/AceBench","type":"external"},{"url":"https://github.com/OpenBMB/Omni-DuplexEval","path":null,"label":"OpenBMB/Omni-DuplexEval","type":"external"},{"url":"https://github.com/OpenBMB/UltraEval","path":null,"label":"OpenBMB/UltraEval","type":"external"},{"url":"https://github.com/OpenBMB/InfiniteBench","path":null,"label":"OpenBMB/InfiniteBench","type":"external"},{"url":"https://github.com/OpenBMB/SHIFT","path":null,"label":"OpenBMB/SHIFT","type":"external"},{"url":"https://github.com/OpenBMB/ForgeTrain","path":null,"label":"OpenBMB/ForgeTrain","type":"external"},{"url":"https://github.com/OpenBMB/VisRAG","path":null,"label":"OpenBMB/VisRAG","type":"external"},{"url":"https://huggingface.co/openbmb/EVisRAG-7B","path":null,"label":"openbmb/EVisRAG-7B","type":"external"},{"url":"https://github.com/OpenBMB/VisCPM","path":null,"label":"OpenBMB/VisCPM","type":"external"},{"url":"https://github.com/OpenBMB/DecT","path":null,"label":"OpenBMB/DecT","type":"external"},{"url":"https://github.com/OpenBMB/General-Model-License","path":null,"label":"OpenBMB/General-Model-License","type":"external"}],"provenance":{"provider":"deepseek","model":"deepseek-v4-pro","workflow":"onlylabs-deepagents-analysis-v3","agent":"deepagents"},"evidence":{"total":93,"pages":28,"events":140,"web":5,"signal_desks":{"forks":2,"repos":12,"hiring":0,"talking":0,"releases":46},"data_radar_lanes":null,"data_radar_matches":null}},"signal_counts":{"total":203,"model_released":31,"release":82,"repo_new":78,"repo_forked":2,"post_published":10,"job_opened":0}}