MiMo-V2.6 MOPD2 checkpoints (Flash + Pro)
Xiaomi MiMo released MOPD2-upgraded MiMo-V2.6-Flash and Pro checkpoints: a multi-teacher on-policy distillation stage that fuses domain-specialized teachers and fixes agentic tool-call repetition.
Why it matters
Multi-teacher on-policy distillation (mixRL teachers for verifiable tasks, SFT teachers for open-domain ones, three rollout streams) extends RL gains to hard-to-verify domains like long-horizon game dev and embodied AI; a 12-step specialized-teacher run eliminated the tool-call repetition loop entirely.
What you could build with it
An agent-tools team could adopt the MOPD recipe (mixRL teachers for verifiable tasks, SFT teachers for open-domain ones) to distill small agentic models that reliably end tool-calling loops; the concrete result is coding agents that finish long-horizon tasks instead of burning budget repeating the same tool call.
Does it hold up?
Too early to judge - the checkpoints landed September 27 and third-party deployments are still limited to independent write-ups; the tool-call repetition fix itself is well documented in Xiaomi's own technical blog.
Built with MiMo-V2.6 MOPD2 checkpoints
- dealignai/MiMo-V2.6-Pro-RL-UNCENSORED documents the tool-call repetition bughf · Community build whose maintainer shipped a patched shard after hitting the exact tool-call repetition loop (hundreds of identical tool calls per turn) that the MOPD2 release addresses.
- dreamax fieldwork notes on MOPD2 and MiMo-V2.6 architecturegithub · Independent research notes summarizing the MOPD2 training recipe and the 309B/15B MoE architecture behind the checkpoints.
Learn more
First spotted on hf: source.
More AI models
ElevenLabs Eleven v4 + v4 Turbo: new emotive TTS architecture with a 100ms real-time variantElevenLabs released Eleven v4 and Eleven v4 Turbo on September 28: a new text-to-speech architecture with inline…models · JEV 0.72VoiceStudio: open-source, fully-local ElevenLabs alternative with voice cloning, dubbing and dictation in 646 languagesVoiceStudio is an open-source, fully-local voice AI studio: voice cloning from a 3-second sample, voice design, video…models · JEV 0.71DeepSeek V4-Flash official API: public beta with upgraded agent capabilities, Responses API and Codex supportDeepSeek's official V4-Flash is now a public beta API with massively upgraded agent capabilities - benchmark scores…models · JEV 0.69Mistral Forge: enterprise platform for training and continuously improving proprietary modelsMistral AI launched Forge (Sep 28) — a full-lifecycle model training platform (pre-training, SFT, DPO/ODPO, RL…models · JEV 0.69kev: open, trainable Jev-like family of small decision models on Qwen3.5/3.8A family of small decision models (0.8B to 27B) built on Qwen3.5/3.8 that reproduces Jev's prefill-only typed-decision…models · JEV 0.68GPT-Synopsys: OpenAI and Synopsys multi-year deal to build an AI model that operates EDA toolsOpenAI and Synopsys signed a multi-year agreement on Sep 30 to jointly develop GPT-Synopsys, a specialized model that…models · JEV 0.66
Get the week's best AI launches, plus 3 ideas worth building
One email every Saturday. Ranked by traction, not hype. Free.