{"schema_version":"onlylabs.public_analysis.v1","url":"https://onlylabs.fyi/analysis/xiaomi","json_url":"https://onlylabs.fyi/analysis/xiaomi/analysis.json","evidence_json_url":"https://onlylabs.fyi/analysis/xiaomi/evidence.json","generated_at":"2026-06-28T02:16:54.578Z","analysis":{"org_slug":"xiaomi","url":"https://onlylabs.fyi/analysis/xiaomi","json_url":"https://onlylabs.fyi/analysis/xiaomi/analysis.json","evidence_json_url":"https://onlylabs.fyi/analysis/xiaomi/evidence.json","dossier_url":"https://onlylabs.fyi/labs/xiaomi","org":{"slug":"xiaomi","name":"Xiaomi (MiMo)","category":"neolab","category_label":"Neolab","homepage_url":"https://mimo.xiaomi.com/"},"title":"Xiaomi (MiMo) analysis","summary":"XiaomiMiMo is executing a full-stack, open-source AI strategy that spans text reasoning, vision, audio, embodied AI, and coding agents — with an accelerating pivot toward the Agent era in mid-2026. The lab's evidence trail reveals a deliberate arc: a reasoning-first 7B model family born from pretraining-to-posttraining optimization; rapid horizontal expansion into vision-language (MiMo-VL, May 2025), audio language…","markdown":"## Thesis\n\nXiaomiMiMo is executing a full-stack, open-source AI strategy that spans text reasoning, vision, audio, embodied AI, and coding agents — with an accelerating pivot toward the Agent era in mid-2026. The lab's evidence trail reveals a deliberate arc: a reasoning-first 7B model family born from pretraining-to-posttraining optimization [P7](https://github.com/XiaomiMiMo/MiMo)[P19](https://huggingface.co/XiaomiMiMo/MiMo-7B-SFT/raw/main/README.md)[E1](https://github.com/XiaomiMiMo/MiMo); rapid horizontal expansion into vision-language (MiMo-VL, May 2025) [P6](https://github.com/XiaomiMiMo/MiMo-VL)[E37](https://github.com/XiaomiMiMo/MiMo-VL), audio language models (MiMo-Audio, September 2025) [P10](https://github.com/XiaomiMiMo/MiMo-Audio)[E36](https://github.com/XiaomiMiMo/MiMo-Audio), and embodied AI (MiMo-Embodied, November 2025) [P13](https://github.com/XiaomiMiMo/MiMo-Embodied)[E38](https://github.com/XiaomiMiMo/MiMo-Embodied); a jump to mixture-of-experts at scale with MiMo-V2-Flash (309B/15B active, December 2025) [P14](https://github.com/XiaomiMiMo/MiMo-V2-Flash)[E31](https://github.com/XiaomiMiMo/MiMo-V2-Flash) and MiMo-V2.5-Pro (1.02T/42B active, April 2026) [E3](https://huggingface.co/XiaomiMiMo/MiMo-V2.5-Pro)[E4](https://huggingface.co/XiaomiMiMo/MiMo-V2.5); and, most recently, a developer-tooling land-grab with MiMo-Code (June 2026) [P4](https://github.com/XiaomiMiMo/MiMo-Code)[E11](https://github.com/XiaomiMiMo/MiMo-Code) and an inference-optimized UltraSpeed serving stack built with TileRT [W1](https://ai-tldr.dev/releases/xiaomi-mimo-v2-5-pro-ultraspeed/)[W2](https://aihola.com/article/xiaomi-mimo-ultraspeed-1000-tokens)[W5](https://mimo.xiaomi.com/blog/mimo-tilert-1000tps). All major artifacts are released under permissive MIT or Apache-2.0 licenses [P4](https://github.com/XiaomiMiMo/MiMo-Code)[P6](https://github.com/XiaomiMiMo/MiMo-VL)[P7](https://github.com/XiaomiMiMo/MiMo)[P10](https://github.com/XiaomiMiMo/MiMo-Audio)[P14](https://github.com/XiaomiMiMo/MiMo-V2-Flash)[E2](https://huggingface.co/XiaomiMiMo/MiMo-V2-Flash)[E3](https://huggingface.co/XiaomiMiMo/MiMo-V2.5-Pro)[E4](https://huggingface.co/XiaomiMiMo/MiMo-V2.5), and the lab has formalized its open-source posture through the Orbit Program — a 100-trillion-token incentive plan for AI builders paired with an Agent Ecosystem Co-construction Plan [W4](https://mimo.mi.com/docs/en-US/news/latest/v2.5-open-sourced). The through-line is a lab betting that self-evolving agents, not static chat models, are the path to AGI [W6](https://www.gate.com/news/detail/xiaomis-ai-model-lead-as-ai-competition-shifts-to-the-agent-era-self-20623061).\n\n## Signal desks\n\n### Hiring\n\n- **Research Scientist – Pre-training** (role_theme: pretraining data, architecture, and scaling) — XiaomiMiMo, no specific team/location cited, sourced from careers page [E24](https://mimo.xiaomi.com/index#joinUs)[E46](https://mimo.xiaomi.com/index#joinUs)\n- **Research Scientist – Post-training** (role_theme: SFT, RL, alignment, reasoning optimization) — XiaomiMiMo, no specific team/location cited [E26](https://mimo.xiaomi.com/index#joinUs)[E47](https://mimo.xiaomi.com/index#joinUs)\n- **Research Scientist – Audio Speech** (role_theme: audio language models, ASR, TTS, spoken dialogue) — XiaomiMiMo, no specific team/location cited [E27](https://mimo.xiaomi.com/index#joinUs)[E48](https://mimo.xiaomi.com/index#joinUs)\n- **Research Scientist – Multimodal** (role_theme: vision-language models, cross-modal alignment) — XiaomiMiMo, no specific team/location cited [E29](https://mimo.xiaomi.com/index#joinUs)[E45](https://mimo.xiaomi.com/index#joinUs)\n- **AI Infrastructure Engineer** (role_theme: inference serving, GPU optimization, cluster engineering) — XiaomiMiMo, no specific team/location cited [E25](https://mimo.xiaomi.com/index#joinUs)[E50](https://mimo.xiaomi.com/index#joinUs)\n- **Knowledge Engineer** (role_theme: knowledge representation, agent memory, data curation — signals agent-native infrastructure buildout) — XiaomiMiMo, no specific team/location cited [E28](https://mimo.xiaomi.com/index#joinUs)[E49](https://mimo.xiaomi.com/index#joinUs)\n\n*Assessment*: Six distinct role themes span the full model lifecycle (pretraining → posttraining → multimodal → audio → infra → knowledge). The Knowledge Engineer role [E28](https://mimo.xiaomi.com/index#joinUs)[E49](https://mimo.xiaomi.com/index#joinUs) is a leading indicator of agent-memory and knowledge-system investment that aligns with MiMo-Code's persistent cross-session memory architecture [P4](https://github.com/XiaomiMiMo/MiMo-Code). No location or team granularity is available in the cited evidence.\n\n### Forks\n\n- **vllm-project/vllm** — forked as XiaomiMiMo/vllm (31 stars, 5 forks, branch: `feat_mimo_mtp_stable_073`) [E43](https://github.com/XiaomiMiMo/vllm)[P15](https://github.com/XiaomiMiMo/vllm). Language: Python. Technical theme: high-throughput LLM inference engine adapted for MiMo-specific Multi-Token Prediction (MTP) serving — directly connects to MiMo-V2-Flash's MTP architecture [P14](https://github.com/XiaomiMiMo/MiMo-V2-Flash) and the MiMo-7B-MTPs model release [E35](https://huggingface.co/XiaomiMiMo/MiMo-7B-MTPs).\n- **EvolvingLMMs-Lab/lmms-eval** — forked as XiaomiMiMo/lmms-eval (71 stars, 5 forks, branch: `mimo_vl_eval`) [E42](https://github.com/XiaomiMiMo/lmms-eval)[P16](https://github.com/XiaomiMiMo/lmms-eval). Language: Python. Technical theme: multimodal evaluation framework extended with a custom `MiVLLM` vLLM-based model wrapper, thinking-VLM protocol adaptation, and embodied/GUI-agent benchmarks — underpins MiMo-VL [P6](https://github.com/XiaomiMiMo/MiMo-VL)[P23](https://huggingface.co/XiaomiMiMo/MiMo-VL-7B-RL/raw/main/README.md) and MiMo-Embodied [P13](https://github.com/XiaomiMiMo/MiMo-Embodied) evaluation pipelines.\n\n*Assessment*: Only two forks in the evidence pack, both highly strategic — inference infrastructure (vllm with MTP) and evaluation infrastructure (lmms-eval for VLMs). Both are directly tied to shipped models [P14](https://github.com/XiaomiMiMo/MiMo-V2-Flash)[P23](https://huggingface.co/XiaomiMiMo/MiMo-VL-7B-RL/raw/main/README.md)[P13](https://github.com/XiaomiMiMo/MiMo-Embodied). No evidence of forks from agent frameworks, data pipelines, or safety tooling.\n\n### Releases\n\n- **MiMo-Code v0.1.0–v0.1.3** (June 10–24, 2026) — open-source terminal-native AI coding agent with cross-session memory (SQLite FTS5), multi-agent architecture (build/plan/compose), multi-provider LLM backend, and MiMo Auto free channel [P1](https://github.com/XiaomiMiMo/MiMo-Code/releases/tag/v0.1.3)[P2](https://github.com/XiaomiMiMo/MiMo-Code/releases/tag/v0.1.2)[P3](https://github.com/XiaomiMiMo/MiMo-Code/releases/tag/v0.1.1)[P4](https://github.com/XiaomiMiMo/MiMo-Code)[P5](https://github.com/XiaomiMiMo/MiMo-Code/releases/tag/v0.1.0)[E11](https://github.com/XiaomiMiMo/MiMo-Code)[E18](https://github.com/XiaomiMiMo/MiMo-Code/releases/tag/v0.1.3)[E19](https://github.com/XiaomiMiMo/MiMo-Code/releases/tag/v0.1.2)[E20](https://github.com/XiaomiMiMo/MiMo-Code/releases/tag/v0.1.1)[E22](https://github.com/XiaomiMiMo/MiMo-Code/releases/tag/v0.1.0). 4,288 GitHub stars, 324 forks, 288 open issues within ~1 day of creation [P4](https://github.com/XiaomiMiMo/MiMo-Code). TypeScript, MIT license [P4](https://github.com/XiaomiMiMo/MiMo-Code).\n- **MiMo-V2.5 series** (April 27, 2026) — MiMo-V2.5 (310B params, 216,867 HF downloads, 332 likes) [E4](https://huggingface.co/XiaomiMiMo/MiMo-V2.5)[E16](https://huggingface.co/XiaomiMiMo/MiMo-V2.5-Base), MiMo-V2.5-Pro (1.02T/42B active, 102,336 downloads, 676 likes) [E3](https://huggingface.co/XiaomiMiMo/MiMo-V2.5-Pro)[E14](https://huggingface.co/XiaomiMiMo/MiMo-V2.5-Pro-Base), MiMo-V2.5-ASR (7.6B, 2,548 downloads, 97 likes) [E9](https://huggingface.co/XiaomiMiMo/MiMo-V2.5-ASR)[E33](https://github.com/XiaomiMiMo/MiMo-V2.5-ASR). All MIT licensed [E3](https://huggingface.co/XiaomiMiMo/MiMo-V2.5-Pro)[E4](https://huggingface.co/XiaomiMiMo/MiMo-V2.5)[E9](https://huggingface.co/XiaomiMiMo/MiMo-V2.5-ASR).\n- **MiMo-V2.5-Pro-UltraSpeed** (June 8, 2026) — inference-optimized FP4-quantized variant of MiMo-V2.5-Pro using Block-Diffusion 'DFlash' speculative decoding, achieving ~1,200 tokens/sec on 8-GPU commodity nodes via TileRT partnership [W1](https://ai-tldr.dev/releases/xiaomi-mimo-v2-5-pro-ultraspeed/)[W2](https://aihola.com/article/xiaomi-mimo-ultraspeed-1000-tokens)[W5](https://mimo.xiaomi.com/blog/mimo-tilert-1000tps). FP4-DFlash checkpoint on HuggingFace, select TileRT modules on GitHub [W1](https://ai-tldr.dev/releases/xiaomi-mimo-v2-5-pro-ultraspeed/)[W3](https://byteiota.com/mimo-ultraspeed-hits-1000-tokens-sec-on-stock-gpus/)[W5](https://mimo.xiaomi.com/blog/mimo-tilert-1000tps). Paid API trial June 9–23 at 3× standard rate [W1](https://ai-tldr.dev/releases/xiaomi-mimo-v2-5-pro-ultraspeed/)[W5](https://mimo.xiaomi.com/blog/mimo-tilert-1000tps).\n- **MiMo-V2-Flash** (December 15, 2025) — 309B total/15B active MoE with hybrid attention (Sliding Window + Global at 5:1 ratio, 128-token window) and Multi-Token Prediction. 68,448 HF downloads, 741 likes [E2](https://huggingface.co/XiaomiMiMo/MiMo-V2-Flash)[P14](https://github.com/XiaomiMiMo/MiMo-V2-Flash)[E31](https://github.com/XiaomiMiMo/MiMo-V2-Flash). 1,333 GitHub stars [E31](https://github.com/XiaomiMiMo/MiMo-V2-Flash).\n- **MiMo-Audio family** (September 18–19, 2025) — MiMo-Audio-7B-Base and Instruct (few-shot audio learners, 100M+ hours pretraining data), MiMo-Audio-Tokenizer (1.2B transformer, 11M hours training, semantic+acoustic unified representation), MiMo-Audio-Eval toolkit, MiMo-Audio-Training toolkit [P8](https://github.com/XiaomiMiMo/MiMo-Audio-Eval)[P9](https://github.com/XiaomiMiMo/MiMo-Audio-Tokenizer)[P10](https://github.com/XiaomiMiMo/MiMo-Audio)[P11](https://github.com/XiaomiMiMo/MiMo-Audio-Training)[E7](https://huggingface.co/XiaomiMiMo/MiMo-Audio-7B-Instruct)[E15](https://huggingface.co/XiaomiMiMo/MiMo-Audio-7B-Base)[E21](https://huggingface.co/XiaomiMiMo/MiMo-Audio-Tokenizer)[E36](https://github.com/XiaomiMiMo/MiMo-Audio)[E39](https://github.com/XiaomiMiMo/MiMo-Audio-Tokenizer)[E40](https://github.com/XiaomiMiMo/MiMo-Audio-Training)[E41](https://github.com/XiaomiMiMo/MiMo-Audio-Eval). MiMo-Audio-7B-Instruct: 24,086 HF downloads, 158 likes [E7](https://huggingface.co/XiaomiMiMo/MiMo-Audio-7B-Instruct). MiMo-Audio repo: 1,046 GitHub stars [E36](https://github.com/XiaomiMiMo/MiMo-Audio).\n- **MiMo-VL family** (May 29–August 7, 2025) — MiMo-VL-7B-SFT and RL, with 2508 update bringing thinking control (`no_think` parameter), MMMU 70.6, VideoMME 70.8 [P6](https://github.com/XiaomiMiMo/MiMo-VL)[P23](https://huggingface.co/XiaomiMiMo/MiMo-VL-7B-RL/raw/main/README.md)[P25](https://huggingface.co/XiaomiMiMo/MiMo-VL-7B-SFT/raw/main/README.md)[P26](https://huggingface.co/XiaomiMiMo/MiMo-VL-7B-SFT-2508/raw/main/README.md)[P27](https://huggingface.co/XiaomiMiMo/MiMo-VL-7B-RL-2508/raw/main/README.md)[E6](https://huggingface.co/XiaomiMiMo/MiMo-VL-7B-RL)[E10](https://huggingface.co/XiaomiMiMo/MiMo-VL-7B-RL-2508)[E13](https://huggingface.co/XiaomiMiMo/MiMo-VL-7B-SFT)[E23](https://huggingface.co/XiaomiMiMo/MiMo-VL-7B-SFT-2508)[E37](https://github.com/XiaomiMiMo/MiMo-VL). 643 GitHub stars [E37](https://github.com/XiaomiMiMo/MiMo-VL).\n- **MiMo-7B reasoning family** (April 26–May 30, 2025) — Base, SFT, RL, RL-Zero, RL-0530. RL-0530 reaches AIME24 80.1 (surpassing DeepSeek R1 at 79.8), MATH500 97.2 [P7](https://github.com/XiaomiMiMo/MiMo)[P19](https://huggingface.co/XiaomiMiMo/MiMo-7B-SFT/raw/main/README.md)[P20](https://huggingface.co/XiaomiMiMo/MiMo-7B-Base/raw/main/README.md)[P21](https://huggingface.co/XiaomiMiMo/MiMo-7B-RL/raw/main/README.md)[P22](https://huggingface.co/XiaomiMiMo/MiMo-7B-RL-Zero/raw/main/README.md)[P24](https://huggingface.co/XiaomiMiMo/MiMo-7B-RL-0530/raw/main/README.md)[E1](https://github.com/XiaomiMiMo/MiMo)[E5](https://huggingface.co/XiaomiMiMo/MiMo-7B-RL)[E8](https://huggingface.co/XiaomiMiMo/MiMo-7B-Base)[E17](https://huggingface.co/XiaomiMiMo/MiMo-7B-RL-0530)[E30](https://huggingface.co/XiaomiMiMo/MiMo-7B-SFT)[E32](https://huggingface.co/XiaomiMiMo/MiMo-7B-RL-Zero). MiMo-7B-RL: 231,513 HF downloads, 277 likes [E5](https://huggingface.co/XiaomiMiMo/MiMo-7B-RL). MiMo repo: 2,167 GitHub stars [E1](https://github.com/XiaomiMiMo/MiMo).\n- **MiMo-Embodied** (November 19, 2025) — cross-embodied VLM for autonomous driving + embodied AI, evaluation suite only [P13](https://github.com/XiaomiMiMo/MiMo-Embodied)[E12](https://huggingface.co/XiaomiMiMo/MiMo-Embodied-7B)[E38](https://github.com/XiaomiMiMo/MiMo-Embodied). MiMo-Embodied-7B: 309 HF downloads, 68 likes [E12](https://huggingface.co/XiaomiMiMo/MiMo-Embodied-7B). 386 GitHub stars [E38](https://github.com/XiaomiMiMo/MiMo-Embodied).\n- **MiMo-Skills** (April 23, 2026) — agent skills package (MiMo V2.5 TTS, voice synthesis/cloning/design), installable via npx, MIT license [P17](https://github.com/XiaomiMiMo/MiMo-Skills)[E34](https://github.com/XiaomiMiMo/MiMo-Skills). 70 GitHub stars [E34](https://github.com/XiaomiMiMo/MiMo-Skills).\n- **MiMo-7B-MTPs** (November 14, 2025) — multi-token prediction model (421M params) [E35](https://huggingface.co/XiaomiMiMo/MiMo-7B-MTPs).\n\n### Talking\n\n- **Agent era and self-evolution narrative** — Xiaomi large-model team lead Luo Fuli gave a 3.5-hour interview (April 24, 2026, Bilibili) arguing the competition track has shifted from the Chat era to the Agent era, and that \"self-evolution\" is the key event on the path to AGI [W6](https://www.gate.com/news/detail/xiaomis-ai-model-lead-as-ai-competition-shifts-to-the-agent-era-self-20623061). This directly frames MiMo-Code's \"Models and Agents Co-Evolve\" tagline [P4](https://github.com/XiaomiMiMo/MiMo-Code)[E11](https://github.com/XiaomiMiMo/MiMo-Code).\n- **MiMo-V2.5 open-source + Orbit Program** — Official announcement releasing MiMo-V2.5 series under MIT license, launching the Orbit 100-trillion-token incentive plan for AI builders and an Agent Ecosystem Co-construction Plan for agent framework teams, with chip manufacturer and inference framework partnerships [W4](https://mimo.mi.com/docs/en-US/news/latest/v2.5-open-sourced).\n- **UltraSpeed: 1,000+ tok/s on commodity hardware** — Multiple outlets covered MiMo-V2.5-Pro-UltraSpeed's Block-Diffusion DFlash speculative decoding partnered with TileRT, emphasizing that the speed runs on stock 8-GPU nodes without custom silicon [W1](https://ai-tldr.dev/releases/xiaomi-mimo-v2-5-pro-ultraspeed/)[W2](https://aihola.com/article/xiaomi-mimo-ultraspeed-1000-tokens)[W3](https://byteiota.com/mimo-ultraspeed-hits-1000-tokens-sec-on-stock-gpus/)[W5](https://mimo.xiaomi.com/blog/mimo-tilert-1000tps). The open-source FP4 checkpoint and DFlash modules are positioned as potentially generalizable beyond Xiaomi's model family [W3](https://byteiota.com/mimo-ultraspeed-hits-1000-tokens-sec-on-stock-gpus/). HN traction: MiMo repo hit 482 points/193 comments at launch [E1](https://github.com/XiaomiMiMo/MiMo); MiMo-V2-Flash repo had 3 points/0 comments [E31](https://github.com/XiaomiMiMo/MiMo-V2-Flash).\n\n## Shipping\n\nMiMo's shipping cadence is relentless and accelerating. The lab released its first public artifact (MiMo-7B reasoning models) in late April 2025 [E1](https://github.com/XiaomiMiMo/MiMo)[E5](https://huggingface.co/XiaomiMiMo/MiMo-7B-RL)[E8](https://huggingface.co/XiaomiMiMo/MiMo-7B-Base)[E30](https://huggingface.co/XiaomiMiMo/MiMo-7B-SFT)[E32](https://huggingface.co/XiaomiMiMo/MiMo-7B-RL-Zero), then shipped vision-language models within one month (May 2025) [E6](https://huggingface.co/XiaomiMiMo/MiMo-VL-7B-RL)[E13](https://huggingface.co/XiaomiMiMo/MiMo-VL-7B-SFT)[E37](https://github.com/XiaomiMiMo/MiMo-VL), audio language models by September 2025 [E7](https://huggingface.co/XiaomiMiMo/MiMo-Audio-7B-Instruct)[E15](https://huggingface.co/XiaomiMiMo/MiMo-Audio-7B-Base)[E36](https://github.com/XiaomiMiMo/MiMo-Audio), embodied AI in November 2025 [E12](https://huggingface.co/XiaomiMiMo/MiMo-Embodied-7B)[E38](https://github.com/XiaomiMiMo/MiMo-Embodied), and a 309B MoE model in December 2025 [E2](https://huggingface.co/XiaomiMiMo/MiMo-V2-Flash)[E31](https://github.com/XiaomiMiMo/MiMo-V2-Flash). The pace intensified in 2026: the MiMo-V2.5 series (including the 1.02T-param Pro variant) landed in late April [E3](https://huggingface.co/XiaomiMiMo/MiMo-V2.5-Pro)[E4](https://huggingface.co/XiaomiMiMo/MiMo-V2.5)[E9](https://huggingface.co/XiaomiMiMo/MiMo-V2.5-ASR)[E14](https://huggingface.co/XiaomiMiMo/MiMo-V2.5-Pro-Base)[E16](https://huggingface.co/XiaomiMiMo/MiMo-V2.5-Base), followed by the MiMo-Code agent in June 2026 [E11](https://github.com/XiaomiMiMo/MiMo-Code)[E18](https://github.com/XiaomiMiMo/MiMo-Code/releases/tag/v0.1.3)[E19](https://github.com/XiaomiMiMo/MiMo-Code/releases/tag/v0.1.2)[E20](https://github.com/XiaomiMiMo/MiMo-Code/releases/tag/v0.1.1)[E22](https://github.com/XiaomiMiMo/MiMo-Code/releases/tag/v0.1.0) and the UltraSpeed inference stack within the same month [W1](https://ai-tldr.dev/releases/xiaomi-mimo-v2-5-pro-ultraspeed/)[W2](https://aihola.com/article/xiaomi-mimo-ultraspeed-1000-tokens)[W5](https://mimo.xiaomi.com/blog/mimo-tilert-1000tps).\n\nThe most strategically significant recent shipment is MiMo-Code [P4](https://github.com/XiaomiMiMo/MiMo-Code)[E11](https://github.com/XiaomiMiMo/MiMo-Code), which bundles a terminal-native coding agent, persistent cross-session memory (SQLite FTS5), three built-in agent modes (build/plan/compose), and a free-for-limited-time MiMo Auto channel — an adoption play targeting the developer-tooling market currently contested by Claude Code, Cursor, and open-source alternatives. The 4,288 stars and 324 forks achieved within ~1 day [P4](https://github.com/XiaomiMiMo/MiMo-Code) suggest substantial launch-day coordination and community interest.\n\nThe UltraSpeed release [W1](https://ai-tldr.dev/releases/xiaomi-mimo-v2-5-pro-ultraspeed/)[W2](https://aihola.com/article/xiaomi-mimo-ultraspeed-1000-tokens)[W5](https://mimo.xiaomi.com/blog/mimo-tilert-1000tps) is a secondary but notable shipment: an FP4-quantized, speculative-decoding-optimized serving mode that achieves ~1,200 tok/s on 8-GPU commodity nodes. The open-sourcing of the FP4-DFlash checkpoint [W1](https://ai-tldr.dev/releases/xiaomi-mimo-v2-5-pro-ultraspeed/)[W3](https://byteiota.com/mimo-ultraspeed-hits-1000-tokens-sec-on-stock-gpus/) positions Xiaomi as contributing inference techniques that other labs could adopt, while the paid API trial (3× standard rate for ~10× speed) [W1](https://ai-tldr.dev/releases/xiaomi-mimo-v2-5-pro-ultraspeed/)[W2](https://aihola.com/article/xiaomi-mimo-ultraspeed-1000-tokens) suggests a commercialization experiment alongside the open release.\n\n## Research themes\n\n1. **Reasoning from pretraining onward**. MiMo-7B was trained from scratch with reasoning-specific pretraining strategies, not merely post-hoc RL on a generic base model [P22](https://huggingface.co/XiaomiMiMo/MiMo-7B-RL-Zero/raw/main/README.md)[P7](https://github.com/XiaomiMiMo/MiMo). The lab explicitly argues that \"the effectiveness of RL-trained reasoning relies on the inherent reasoning potential of the base model\" [P22](https://huggingface.co/XiaomiMiMo/MiMo-7B-RL-Zero/raw/main/README.md). Scaling SFT data from 500K to 6M instances and expanding RL context windows from 32K to 48K produced MiMo-7B-RL-0530, which surpasses DeepSeek R1 on AIME24 (80.1 vs 79.8) [P7](https://github.com/XiaomiMiMo/MiMo)[P19](https://huggingface.co/XiaomiMiMo/MiMo-7B-SFT/raw/main/README.md)[P24](https://huggingface.co/XiaomiMiMo/MiMo-7B-RL-0530/raw/main/README.md).\n\n2. **Mixed On-policy Reinforcement Learning (MORL) for VLMs**. MiMo-VL introduces MORL, a framework integrating diverse reward signals spanning perception accuracy, visual grounding, logical reasoning, and human/AI preferences into a single RL post-training stage [P23](https://huggingface.co/XiaomiMiMo/MiMo-VL-7B-RL/raw/main/README.md)[P25](https://huggingface.co/XiaomiMiMo/MiMo-VL-7B-SFT/raw/main/README.md).\n\n3. **Few-shot audio language models**. MiMo-Audio scales pretraining to \"over one hundred million hours\" to elicit few-shot generalization across audio tasks without task-specific fine-tuning — an explicit parallel to GPT-3's text paradigm applied to audio [P10](https://github.com/XiaomiMiMo/MiMo-Audio)[P28](https://huggingface.co/XiaomiMiMo/MiMo-Audio-7B-Base/raw/main/README.md).\n\n4. **Hybrid attention for efficient long-context MoE**. MiMo-V2-Flash interleaves Sliding Window Attention and Global Attention at a 5:1 ratio with a 128-token window and learnable attention sink bias, reducing KV-cache storage by ~6× while maintaining long-context performance [P14](https://github.com/XiaomiMiMo/MiMo-V2-Flash).\n\n5. **Unified audio tokenization**. MiMo-Audio-Tokenizer is a 1.2B pure-transformer trained from scratch on 11M hours, jointly handling semantic extraction and high-fidelity reconstruction — aiming to resolve the semantic-acoustic representation conflict [P9](https://github.com/XiaomiMiMo/MiMo-Audio-Tokenizer).\n\n6. **Cross-embodied VLMs**. MiMo-Embodied targets both autonomous driving and embodied AI tasks in a single VLM, described as \"the first open-source VLM that integrates these two critical areas\" [P13](https://github.com/XiaomiMiMo/MiMo-Embodied).\n\n7. **Agent self-evolution**. The lab's public narrative [W6](https://www.gate.com/news/detail/xiaomis-ai-model-lead-as-ai-competition-shifts-to-the-agent-era-self-20623061) and MiMo-Code's architecture [P4](https://github.com/XiaomiMiMo/MiMo-Code) point to an emerging research theme around agents that improve across sessions — MiMo-Code's persistent memory system (MEMORY.md with SQLite FTS5) and its framing as \"Where Models and Agents Co-Evolve\" [E11](https://github.com/XiaomiMiMo/MiMo-Code) signal applied research into agentic self-improvement loops.\n\n## Hiring & scaling\n\nSix role categories are open simultaneously on the MiMo careers page , covering the full model-development pipeline. The pattern reveals a lab scaling on multiple fronts:\n\n- **Core model R&D**: Pre-training [E24](https://mimo.xiaomi.com/index#joinUs)[E46](https://mimo.xiaomi.com/index#joinUs) and Post-training [E26](https://mimo.xiaomi.com/index#joinUs)[E47](https://mimo.xiaomi.com/index#joinUs) roles indicate continued investment in foundation model improvement, consistent with the trajectory from MiMo-7B to MiMo-V2.5-Pro.\n- **Modality expansion**: Audio Speech [E27](https://mimo.xiaomi.com/index#joinUs)[E48](https://mimo.xiaomi.com/index#joinUs) and Multimodal [E29](https://mimo.xiaomi.com/index#joinUs)[E45](https://mimo.xiaomi.com/index#joinUs) roles confirm that audio and vision are not one-off projects but sustained research programs.\n- **Agent infrastructure**: The Knowledge Engineer role [E28](https://mimo.xiaomi.com/index#joinUs)[E49](https://mimo.xiaomi.com/index#joinUs) is a differentiating signal — it implies investment in knowledge representation, retrieval, and memory systems that underpin agent persistence [P4](https://github.com/XiaomiMiMo/MiMo-Code). The AI Infrastructure Engineer role [E25](https://mimo.xiaomi.com/index#joinUs)[E50](https://mimo.xiaomi.com/index#joinUs) points to inference-serving and cluster-engineering needs that align with the UltraSpeed push [W1](https://ai-tldr.dev/releases/xiaomi-mimo-v2-5-pro-ultraspeed/)[W2](https://aihola.com/article/xiaomi-mimo-ultraspeed-1000-tokens)[W5](https://mimo.xiaomi.com/blog/mimo-tilert-1000tps) and the Orbit program's chip-manufacturer collaborations [W4](https://mimo.mi.com/docs/en-US/news/latest/v2.5-open-sourced).\n\n**Evidence gap**: No cited evidence provides headcount, location, team size, compensation, or whether roles are net-new or backfill. The careers page URL (mimo.xiaomi.com/index#joinUs) is the sole source. The absence of product management, developer relations, or GTM roles in the evidence pack may reflect scope of collection rather than actual hiring posture.\n\n## Category implications\n\n**Strategy**: XiaomiMiMo is pursuing a platform strategy disguised as a model lab. The Orbit Program — combining a 100-trillion-token incentive for builders and an Agent Ecosystem Co-construction Plan for framework teams [W4](https://mimo.mi.com/docs/en-US/news/latest/v2.5-open-sourced) — positions MiMo models as infrastructure that third-party developers and chip manufacturers adopt, not merely consume. MIT licensing across flagship models [E2](https://huggingface.co/XiaomiMiMo/MiMo-V2-Flash)[E3](https://huggingface.co/XiaomiMiMo/MiMo-V2.5-Pro)[E4](https://huggingface.co/XiaomiMiMo/MiMo-V2.5)[E5](https://huggingface.co/XiaomiMiMo/MiMo-7B-RL) removes friction for commercial uptake. The MiMo-Code agent [P4](https://github.com/XiaomiMiMo/MiMo-Code) and MiMo-Skills package [P17](https://github.com/XiaomiMiMo/MiMo-Skills) extend this platform play into developer tooling, where distribution (npm, npx, one-line curl install) and multi-provider compatibility lower switching costs from incumbent coding agents.\n\n**Infrastructure**: The vllm fork with MTP support [E43](https://github.com/XiaomiMiMo/vllm)[P15](https://github.com/XiaomiMiMo/vllm) and the TileRT partnership for UltraSpeed [W1](https://ai-tldr.dev/releases/xiaomi-mimo-v2-5-pro-ultraspeed/)[W2](https://aihola.com/article/xiaomi-mimo-ultraspeed-1000-tokens)[W5](https://mimo.xiaomi.com/blog/mimo-tilert-1000tps) indicate that inference efficiency is treated as a first-class research problem, not an afterthought. The FP4 quantization and speculative decoding technique is being open-sourced in a generalizable form [W3](https://byteiota.com/mimo-ultraspeed-hits-1000-tokens-sec-on-stock-gpus/), which could influence inference norms beyond Xiaomi's own model family. Hiring for AI Infrastructure Engineers [E25](https://mimo.xiaomi.com/index#joinUs)[E50](https://mimo.xiaomi.com/index#joinUs) confirms this is a build (not buy) function.\n\n**Product**: MiMo-Code [P4](https://github.com/XiaomiMiMo/MiMo-Code) is the clearest product signal — a terminal-native coding agent with persistent memory, multi-agent routing, and a free-tier acquisition channel. The MiMo-Skills repo (starting with TTS) [P17](https://github.com/XiaomiMiMo/MiMo-Skills) suggests a plugin/extension model for agent capabilities. MiMo-V2.5-ASR [P18](https://github.com/XiaomiMiMo/MiMo-V2.5-ASR)[E9](https://huggingface.co/XiaomiMiMo/MiMo-V2.5-ASR) targets production speech-recognition use cases (multilingual, dialect, code-switching, noisy environments). The MiMo-V2.5-Pro-UltraSpeed paid API trial [W1](https://ai-tldr.dev/releases/xiaomi-mimo-v2-5-pro-ultraspeed/)[W5](https://mimo.xiaomi.com/blog/mimo-tilert-1000tps) is an explicit commercialization experiment, testing willingness-to-pay for inference speed.\n\n**Research**: The lab's research program is unusually broad for its apparent size — spanning reasoning, vision, audio, embodied AI, and agent self-evolution simultaneously. The unifying thesis appears to be that reasoning potential is built at pretraining time [P22](https://huggingface.co/XiaomiMiMo/MiMo-7B-RL-Zero/raw/main/README.md) and can be extended across modalities (VL, audio, embodied) through scaling and RL. The MORL framework [P23](https://huggingface.co/XiaomiMiMo/MiMo-VL-7B-RL/raw/main/README.md) and few-shot audio paradigm [P10](https://github.com/XiaomiMiMo/MiMo-Audio) are novel contributions. The \"self-evolution\" framing [W6](https://www.gate.com/news/detail/xiaomis-ai-model-lead-as-ai-competition-shifts-to-the-agent-era-self-20623061) suggests the next research frontier is agentic loops where models improve through interaction, not static training.\n\n**Hiring implications**: The six concurrent role categories suggest headcount growth across the stack. Competitors should monitor whether the Knowledge Engineer [E28](https://mimo.xiaomi.com/index#joinUs)[E49](https://mimo.xiaomi.com/index#joinUs) and AI Infrastructure Engineer [E25](https://mimo.xiaomi.com/index#joinUs)[E50](https://mimo.xiaomi.com/index#joinUs) roles grow in volume — they are leading indicators of agent-memory and serving-infrastructure buildout, respectively.\n\n**GTM implications**: MiMo is using open-source as its primary GTM motion — MIT-licensed weights [E2](https://huggingface.co/XiaomiMiMo/MiMo-V2-Flash)[E3](https://huggingface.co/XiaomiMiMo/MiMo-V2.5-Pro)[E4](https://huggingface.co/XiaomiMiMo/MiMo-V2.5)[E5](https://huggingface.co/XiaomiMiMo/MiMo-7B-RL), open evaluation toolkits [P8](https://github.com/XiaomiMiMo/MiMo-Audio-Eval)[P16](https://github.com/XiaomiMiMo/lmms-eval), open training toolkits [P11](https://github.com/XiaomiMiMo/MiMo-Audio-Training), and open inference code [W1](https://ai-tldr.dev/releases/xiaomi-mimo-v2-5-pro-ultraspeed/)[W3](https://byteiota.com/mimo-ultraspeed-hits-1000-tokens-sec-on-stock-gpus/). The MiMo Auto free channel in MiMo-Code [P4](https://github.com/XiaomiMiMo/MiMo-Code) and the Orbit token incentive [W4](https://mimo.mi.com/docs/en-US/news/latest/v2.5-open-sourced) are demand-generation tactics. The API platform (platform.xiaomimimo.com) [P14](https://github.com/XiaomiMiMo/MiMo-V2-Flash)[P17](https://github.com/XiaomiMiMo/MiMo-Skills) and MiMo Studio [P14](https://github.com/XiaomiMiMo/MiMo-V2-Flash) suggest a parallel hosted revenue model. The 3× pricing on UltraSpeed [W1](https://ai-tldr.dev/releases/xiaomi-mimo-v2-5-pro-ultraspeed/)[W2](https://aihola.com/article/xiaomi-mimo-ultraspeed-1000-tokens) tests price discrimination by inference latency.\n\n## Traction highlights\n\n- **MiMo-Code**: 4,288 GitHub stars, 324 forks, 288 open issues within ~1 day of repository creation (June 10–11, 2026) [P4](https://github.com/XiaomiMiMo/MiMo-Code). Event-level data records 10,885 stars shortly after [E11](https://github.com/XiaomiMiMo/MiMo-Code). Four patch releases in 14 days (v0.1.0 → v0.1.3, June 10–24, 2026) [P1](https://github.com/XiaomiMiMo/MiMo-Code/releases/tag/v0.1.3)[P2](https://github.com/XiaomiMiMo/MiMo-Code/releases/tag/v0.1.2)[P3](https://github.com/XiaomiMiMo/MiMo-Code/releases/tag/v0.1.1)[P5](https://github.com/XiaomiMiMo/MiMo-Code/releases/tag/v0.1.0).\n- **MiMo-V2.5**: 216,867 HuggingFace downloads, 332 likes [E4](https://huggingface.co/XiaomiMiMo/MiMo-V2.5). MiMo-V2.5-Pro: 102,336 downloads, 676 likes [E3](https://huggingface.co/XiaomiMiMo/MiMo-V2.5-Pro).\n- **MiMo-7B-RL**: 231,513 HuggingFace downloads, 277 likes [E5](https://huggingface.co/XiaomiMiMo/MiMo-7B-RL). MiMo-7B-Base: 186,271 downloads, 135 likes [E8](https://huggingface.co/XiaomiMiMo/MiMo-7B-Base).\n- **MiMo-V2-Flash**: 68,448 HuggingFace downloads, 741 likes [E2](https://huggingface.co/XiaomiMiMo/MiMo-V2-Flash); 1,333 GitHub stars [E31](https://github.com/XiaomiMiMo/MiMo-V2-Flash).\n- **MiMo-Audio-7B-Instruct**: 24,086 HuggingFace downloads, 158 likes [E7](https://huggingface.co/XiaomiMiMo/MiMo-Audio-7B-Instruct). MiMo-Audio GitHub repo: 1,046 stars [E36](https://github.com/XiaomiMiMo/MiMo-Audio).\n- **MiMo (core reasoning)**: 2,167 GitHub stars; HN launch traction of 482 points/193 comments [E1](https://github.com/XiaomiMiMo/MiMo).\n- **MiMo-VL**: 643 GitHub stars [E37](https://github.com/XiaomiMiMo/MiMo-VL).\n- **MiMo-Embodied**: 386 GitHub stars [E38](https://github.com/XiaomiMiMo/MiMo-Embodied).\n- **MiMo-V2.5-ASR**: 264 GitHub stars [E33](https://github.com/XiaomiMiMo/MiMo-V2.5-ASR).\n\n*Note on measurement inconsistency*: GitHub star counts vary between page-scrape evidence [P4](https://github.com/XiaomiMiMo/MiMo-Code)[P6](https://github.com/XiaomiMiMo/MiMo-VL)[P7](https://github.com/XiaomiMiMo/MiMo)[P10](https://github.com/XiaomiMiMo/MiMo-Audio)[P13](https://github.com/XiaomiMiMo/MiMo-Embodied)[P14](https://github.com/XiaomiMiMo/MiMo-V2-Flash)[P17](https://github.com/XiaomiMiMo/MiMo-Skills)[P18](https://github.com/XiaomiMiMo/MiMo-V2.5-ASR) and event-level evidence [E1](https://github.com/XiaomiMiMo/MiMo)[E11](https://github.com/XiaomiMiMo/MiMo-Code)[E31](https://github.com/XiaomiMiMo/MiMo-V2-Flash)[E33](https://github.com/XiaomiMiMo/MiMo-V2.5-ASR)[E34](https://github.com/XiaomiMiMo/MiMo-Skills)[E36](https://github.com/XiaomiMiMo/MiMo-Audio)[E37](https://github.com/XiaomiMiMo/MiMo-VL)[E38](https://github.com/XiaomiMiMo/MiMo-Embodied)[E39](https://github.com/XiaomiMiMo/MiMo-Audio-Tokenizer)[E40](https://github.com/XiaomiMiMo/MiMo-Audio-Training)[E41](https://github.com/XiaomiMiMo/MiMo-Audio-Eval) due to different capture times. Both are cited where available; event-level data generally postdates page-level data.\n\n## Sources\n\nP1–P28: Repository pages, release pages, model cards, and README files from GitHub and HuggingFace. E1–E50: Structured event records (repo creation, model releases, job postings, fork events). W1–W6: Web articles and blog posts.","generated_at":"2026-06-27T22:26:08.312+00:00","citations":[{"url":"https://github.com/XiaomiMiMo/MiMo","path":null,"label":"XiaomiMiMo/MiMo","type":"external"},{"url":"https://huggingface.co/XiaomiMiMo/MiMo-7B-SFT/raw/main/README.md","path":null,"label":"XiaomiMiMo/MiMo-7B-SFT","type":"external"},{"url":"https://github.com/XiaomiMiMo/MiMo","path":null,"label":"XiaomiMiMo/MiMo","type":"external"},{"url":"https://github.com/XiaomiMiMo/MiMo-VL","path":null,"label":"XiaomiMiMo/MiMo-VL","type":"external"},{"url":"https://github.com/XiaomiMiMo/MiMo-VL","path":null,"label":"XiaomiMiMo/MiMo-VL","type":"external"},{"url":"https://github.com/XiaomiMiMo/MiMo-Audio","path":null,"label":"XiaomiMiMo/MiMo-Audio","type":"external"},{"url":"https://github.com/XiaomiMiMo/MiMo-Audio","path":null,"label":"XiaomiMiMo/MiMo-Audio","type":"external"},{"url":"https://github.com/XiaomiMiMo/MiMo-Embodied","path":null,"label":"XiaomiMiMo/MiMo-Embodied","type":"external"},{"url":"https://github.com/XiaomiMiMo/MiMo-Embodied","path":null,"label":"XiaomiMiMo/MiMo-Embodied","type":"external"},{"url":"https://github.com/XiaomiMiMo/MiMo-V2-Flash","path":null,"label":"XiaomiMiMo/MiMo-V2-Flash","type":"external"},{"url":"https://github.com/XiaomiMiMo/MiMo-V2-Flash","path":null,"label":"XiaomiMiMo/MiMo-V2-Flash","type":"external"},{"url":"https://huggingface.co/XiaomiMiMo/MiMo-V2.5-Pro","path":null,"label":"XiaomiMiMo/MiMo-V2.5-Pro","type":"external"},{"url":"https://huggingface.co/XiaomiMiMo/MiMo-V2.5","path":null,"label":"XiaomiMiMo/MiMo-V2.5","type":"external"},{"url":"https://github.com/XiaomiMiMo/MiMo-Code","path":null,"label":"XiaomiMiMo/MiMo-Code","type":"external"},{"url":"https://github.com/XiaomiMiMo/MiMo-Code","path":null,"label":"XiaomiMiMo/MiMo-Code","type":"external"},{"url":"https://ai-tldr.dev/releases/xiaomi-mimo-v2-5-pro-ultraspeed/","path":null,"label":"ai-tldr.dev/releases","type":"external"},{"url":"https://aihola.com/article/xiaomi-mimo-ultraspeed-1000-tokens","path":null,"label":"aihola.com/article","type":"external"},{"url":"https://mimo.xiaomi.com/blog/mimo-tilert-1000tps","path":null,"label":"mimo.xiaomi.com/blog","type":"external"},{"url":"https://huggingface.co/XiaomiMiMo/MiMo-V2-Flash","path":null,"label":"XiaomiMiMo/MiMo-V2-Flash","type":"external"},{"url":"https://mimo.mi.com/docs/en-US/news/latest/v2.5-open-sourced","path":null,"label":"mimo.mi.com/docs","type":"external"},{"url":"https://www.gate.com/news/detail/xiaomis-ai-model-lead-as-ai-competition-shifts-to-the-agent-era-self-20623061","path":null,"label":"gate.com/news","type":"external"},{"url":"https://mimo.xiaomi.com/index#joinUs","path":null,"label":"mimo.xiaomi.com/index","type":"external"},{"url":"https://mimo.xiaomi.com/index#joinUs","path":null,"label":"mimo.xiaomi.com/index","type":"external"},{"url":"https://mimo.xiaomi.com/index#joinUs","path":null,"label":"mimo.xiaomi.com/index","type":"external"},{"url":"https://mimo.xiaomi.com/index#joinUs","path":null,"label":"mimo.xiaomi.com/index","type":"external"},{"url":"https://mimo.xiaomi.com/index#joinUs","path":null,"label":"mimo.xiaomi.com/index","type":"external"},{"url":"https://mimo.xiaomi.com/index#joinUs","path":null,"label":"mimo.xiaomi.com/index","type":"external"},{"url":"https://mimo.xiaomi.com/index#joinUs","path":null,"label":"mimo.xiaomi.com/index","type":"external"},{"url":"https://mimo.xiaomi.com/index#joinUs","path":null,"label":"mimo.xiaomi.com/index","type":"external"},{"url":"https://mimo.xiaomi.com/index#joinUs","path":null,"label":"mimo.xiaomi.com/index","type":"external"},{"url":"https://mimo.xiaomi.com/index#joinUs","path":null,"label":"mimo.xiaomi.com/index","type":"external"},{"url":"https://mimo.xiaomi.com/index#joinUs","path":null,"label":"mimo.xiaomi.com/index","type":"external"},{"url":"https://mimo.xiaomi.com/index#joinUs","path":null,"label":"mimo.xiaomi.com/index","type":"external"},{"url":"https://github.com/XiaomiMiMo/vllm","path":null,"label":"XiaomiMiMo/vllm","type":"external"},{"url":"https://github.com/XiaomiMiMo/vllm","path":null,"label":"XiaomiMiMo/vllm","type":"external"},{"url":"https://huggingface.co/XiaomiMiMo/MiMo-7B-MTPs","path":null,"label":"XiaomiMiMo/MiMo-7B-MTPs","type":"external"},{"url":"https://github.com/XiaomiMiMo/lmms-eval","path":null,"label":"XiaomiMiMo/lmms-eval","type":"external"},{"url":"https://github.com/XiaomiMiMo/lmms-eval","path":null,"label":"XiaomiMiMo/lmms-eval","type":"external"},{"url":"https://huggingface.co/XiaomiMiMo/MiMo-VL-7B-RL/raw/main/README.md","path":null,"label":"XiaomiMiMo/MiMo-VL-7B-RL","type":"external"},{"url":"https://github.com/XiaomiMiMo/MiMo-Code/releases/tag/v0.1.3","path":null,"label":"XiaomiMiMo/MiMo-Code","type":"external"},{"url":"https://github.com/XiaomiMiMo/MiMo-Code/releases/tag/v0.1.2","path":null,"label":"XiaomiMiMo/MiMo-Code","type":"external"},{"url":"https://github.com/XiaomiMiMo/MiMo-Code/releases/tag/v0.1.1","path":null,"label":"XiaomiMiMo/MiMo-Code","type":"external"},{"url":"https://github.com/XiaomiMiMo/MiMo-Code/releases/tag/v0.1.0","path":null,"label":"XiaomiMiMo/MiMo-Code","type":"external"},{"url":"https://github.com/XiaomiMiMo/MiMo-Code/releases/tag/v0.1.3","path":null,"label":"XiaomiMiMo/MiMo-Code","type":"external"},{"url":"https://github.com/XiaomiMiMo/MiMo-Code/releases/tag/v0.1.2","path":null,"label":"XiaomiMiMo/MiMo-Code","type":"external"},{"url":"https://github.com/XiaomiMiMo/MiMo-Code/releases/tag/v0.1.1","path":null,"label":"XiaomiMiMo/MiMo-Code","type":"external"},{"url":"https://github.com/XiaomiMiMo/MiMo-Code/releases/tag/v0.1.0","path":null,"label":"XiaomiMiMo/MiMo-Code","type":"external"},{"url":"https://huggingface.co/XiaomiMiMo/MiMo-V2.5-Base","path":null,"label":"XiaomiMiMo/MiMo-V2.5-Base","type":"external"},{"url":"https://huggingface.co/XiaomiMiMo/MiMo-V2.5-Pro-Base","path":null,"label":"XiaomiMiMo/MiMo-V2.5-Pro-Base","type":"external"},{"url":"https://huggingface.co/XiaomiMiMo/MiMo-V2.5-ASR","path":null,"label":"XiaomiMiMo/MiMo-V2.5-ASR","type":"external"},{"url":"https://github.com/XiaomiMiMo/MiMo-V2.5-ASR","path":null,"label":"XiaomiMiMo/MiMo-V2.5-ASR","type":"external"},{"url":"https://byteiota.com/mimo-ultraspeed-hits-1000-tokens-sec-on-stock-gpus/","path":null,"label":"byteiota.com/mimo-ultraspeed-hits-1000-tokens-sec-on-stock-gpus","type":"external"},{"url":"https://github.com/XiaomiMiMo/MiMo-Audio-Eval","path":null,"label":"XiaomiMiMo/MiMo-Audio-Eval","type":"external"},{"url":"https://github.com/XiaomiMiMo/MiMo-Audio-Tokenizer","path":null,"label":"XiaomiMiMo/MiMo-Audio-Tokenizer","type":"external"},{"url":"https://github.com/XiaomiMiMo/MiMo-Audio-Training","path":null,"label":"XiaomiMiMo/MiMo-Audio-Training","type":"external"},{"url":"https://huggingface.co/XiaomiMiMo/MiMo-Audio-7B-Instruct","path":null,"label":"XiaomiMiMo/MiMo-Audio-7B-Instruct","type":"external"},{"url":"https://huggingface.co/XiaomiMiMo/MiMo-Audio-7B-Base","path":null,"label":"XiaomiMiMo/MiMo-Audio-7B-Base","type":"external"},{"url":"https://huggingface.co/XiaomiMiMo/MiMo-Audio-Tokenizer","path":null,"label":"XiaomiMiMo/MiMo-Audio-Tokenizer","type":"external"},{"url":"https://github.com/XiaomiMiMo/MiMo-Audio-Tokenizer","path":null,"label":"XiaomiMiMo/MiMo-Audio-Tokenizer","type":"external"},{"url":"https://github.com/XiaomiMiMo/MiMo-Audio-Training","path":null,"label":"XiaomiMiMo/MiMo-Audio-Training","type":"external"},{"url":"https://github.com/XiaomiMiMo/MiMo-Audio-Eval","path":null,"label":"XiaomiMiMo/MiMo-Audio-Eval","type":"external"},{"url":"https://huggingface.co/XiaomiMiMo/MiMo-VL-7B-SFT/raw/main/README.md","path":null,"label":"XiaomiMiMo/MiMo-VL-7B-SFT","type":"external"},{"url":"https://huggingface.co/XiaomiMiMo/MiMo-VL-7B-SFT-2508/raw/main/README.md","path":null,"label":"XiaomiMiMo/MiMo-VL-7B-SFT-2508","type":"external"},{"url":"https://huggingface.co/XiaomiMiMo/MiMo-VL-7B-RL-2508/raw/main/README.md","path":null,"label":"XiaomiMiMo/MiMo-VL-7B-RL-2508","type":"external"},{"url":"https://huggingface.co/XiaomiMiMo/MiMo-VL-7B-RL","path":null,"label":"XiaomiMiMo/MiMo-VL-7B-RL","type":"external"},{"url":"https://huggingface.co/XiaomiMiMo/MiMo-VL-7B-RL-2508","path":null,"label":"XiaomiMiMo/MiMo-VL-7B-RL-2508","type":"external"},{"url":"https://huggingface.co/XiaomiMiMo/MiMo-VL-7B-SFT","path":null,"label":"XiaomiMiMo/MiMo-VL-7B-SFT","type":"external"},{"url":"https://huggingface.co/XiaomiMiMo/MiMo-VL-7B-SFT-2508","path":null,"label":"XiaomiMiMo/MiMo-VL-7B-SFT-2508","type":"external"},{"url":"https://huggingface.co/XiaomiMiMo/MiMo-7B-Base/raw/main/README.md","path":null,"label":"XiaomiMiMo/MiMo-7B-Base","type":"external"},{"url":"https://huggingface.co/XiaomiMiMo/MiMo-7B-RL/raw/main/README.md","path":null,"label":"XiaomiMiMo/MiMo-7B-RL","type":"external"},{"url":"https://huggingface.co/XiaomiMiMo/MiMo-7B-RL-Zero/raw/main/README.md","path":null,"label":"XiaomiMiMo/MiMo-7B-RL-Zero","type":"external"},{"url":"https://huggingface.co/XiaomiMiMo/MiMo-7B-RL-0530/raw/main/README.md","path":null,"label":"XiaomiMiMo/MiMo-7B-RL-0530","type":"external"},{"url":"https://huggingface.co/XiaomiMiMo/MiMo-7B-RL","path":null,"label":"XiaomiMiMo/MiMo-7B-RL","type":"external"},{"url":"https://huggingface.co/XiaomiMiMo/MiMo-7B-Base","path":null,"label":"XiaomiMiMo/MiMo-7B-Base","type":"external"},{"url":"https://huggingface.co/XiaomiMiMo/MiMo-7B-RL-0530","path":null,"label":"XiaomiMiMo/MiMo-7B-RL-0530","type":"external"},{"url":"https://huggingface.co/XiaomiMiMo/MiMo-7B-SFT","path":null,"label":"XiaomiMiMo/MiMo-7B-SFT","type":"external"},{"url":"https://huggingface.co/XiaomiMiMo/MiMo-7B-RL-Zero","path":null,"label":"XiaomiMiMo/MiMo-7B-RL-Zero","type":"external"},{"url":"https://huggingface.co/XiaomiMiMo/MiMo-Embodied-7B","path":null,"label":"XiaomiMiMo/MiMo-Embodied-7B","type":"external"},{"url":"https://github.com/XiaomiMiMo/MiMo-Skills","path":null,"label":"XiaomiMiMo/MiMo-Skills","type":"external"},{"url":"https://github.com/XiaomiMiMo/MiMo-Skills","path":null,"label":"XiaomiMiMo/MiMo-Skills","type":"external"},{"url":"https://huggingface.co/XiaomiMiMo/MiMo-Audio-7B-Base/raw/main/README.md","path":null,"label":"XiaomiMiMo/MiMo-Audio-7B-Base","type":"external"},{"url":"https://github.com/XiaomiMiMo/MiMo-V2.5-ASR","path":null,"label":"XiaomiMiMo/MiMo-V2.5-ASR","type":"external"}],"provenance":{"provider":"deepseek","model":"deepseek-v4-pro","workflow":"onlylabs-deepagents-analysis-v3","agent":"deepagents"},"evidence":{"total":84,"pages":28,"events":50,"web":6,"signal_desks":{"forks":2,"repos":12,"hiring":12,"talking":0,"releases":24},"data_radar_lanes":null,"data_radar_matches":null}},"signal_counts":{"total":50,"model_released":20,"release":4,"repo_new":12,"repo_forked":2,"post_published":0,"job_opened":12}}