{"version":1,"type":"story","url":"https://digestai.news/story/xiaomi-releases-mimo-v2-6-pro-rl-with-1m-token-context-and-1-02t-param","json":"https://digestai.news/story/xiaomi-releases-mimo-v2-6-pro-rl-with-1m-token-context-and-1-02t-param.json","markdown":"https://digestai.news/story/xiaomi-releases-mimo-v2-6-pro-rl-with-1m-token-context-and-1-02t-param.md","slug":"xiaomi-releases-mimo-v2-6-pro-rl-with-1m-token-context-and-1-02t-param","headline":"Xiaomi releases MiMo-V2.6-Pro-RL with 1M-token context and 1.02T parameters","summary":"Xiaomi introduced MiMo-V2.6-Pro-RL, the flagship checkpoint of the MiMo-V2.6 series, built to scale reinforcement learning toward self‑improvement. The model is native omnimodal, handling text, image, video, and audio, and supports a 1M‑token context window for long‑horizon tasks.\n\nThe architecture uses a Sparse Mixture of Experts with 1.02 T total parameters and 42 B activated parameters. It includes a 681M‑param MiMo ViT vision encoder, a 308M AudioTokenizer plus a 127M audio patch encoder, and a 5‑layer speculative decoder. Training employed Group Relative Policy Optimization on very large batches of 1,568 prompts × 16 rollouts per step, with billions of tokens per update, and a groupwise agentic grading loop that ranks rollouts to drive self‑improvement. Deployment instructions reference SGLang and vLLM Docker images.","keyPoints":["MiMo-V2.6-Pro-RL supports text, image, video, audio with 1 M token context length.","Model has 1.02 T total parameters, 42 B activated, using Sparse MoE architecture.","RL training uses 1,568 prompts × 16 rollouts per step and groupwise agentic grading for self‑improvement."],"whyItMatters":"A trillion‑parameter, multimodal model that integrates large‑scale reinforcement learning and self‑improvement loops could accelerate development of autonomous agents and complex task solving.","category":{"slug":"models","name":"Generative AI & Models","url":"https://digestai.news/category/models"},"entities":{"companies":["Xiaomi","lmsysorg","vllm"],"models":["MiMo-V2.6-Pro-RL"],"people":[]},"firstPublishedAt":"2026-09-21T20:06:48Z","updatedAt":"2026-09-21T20:51:40Z","sourceCount":3,"hasPrimarySource":true,"sources":[{"outlet":"huggingface.co","title":"XiaomiMiMo/MiMo-V2.6-Pro-RL · Hugging Face","url":"https://huggingface.co/XiaomiMiMo/MiMo-V2.6-Pro-RL","publishedAt":"2026-09-21T20:06:48Z","type":"primary","primary":true,"lead":true},{"outlet":"huggingface.co","title":"XiaomiMiMo/MiMo-V2.6-Distill-Qwen-9B","url":"https://huggingface.co/XiaomiMiMo/MiMo-V2.6-Distill-Qwen-9B","publishedAt":"2026-09-21T20:51:40Z","type":"primary","primary":true,"lead":false},{"outlet":"huggingface.co","title":"XiaomiMiMo/MiMo-V2.6-Flash-RL · Hugging Face","url":"https://huggingface.co/XiaomiMiMo/MiMo-V2.6-Flash-RL","publishedAt":"2026-09-21T20:09:12Z","type":"primary","primary":true,"lead":false}],"sourceNotes":null,"discussions":[{"site":"Reddit","url":"https://www.reddit.com/r/LocalLLaMA/comments/1wmp2di/xiaomimimomimov26distillqwen9b/","points":null},{"site":"Reddit","url":"https://www.reddit.com/r/LocalLLaMA/comments/1wmnw89/xiaomimimomimov26flashrl_hugging_face/","points":null},{"site":"Reddit","url":"https://www.reddit.com/r/LocalLLaMA/comments/1wmntwv/xiaomimimomimov26prorl_hugging_face/","points":null}],"thread":null,"cite":{"text":"Digest AI, \"Xiaomi releases MiMo-V2.6-Pro-RL with 1M-token context and 1.02T parameters\", 21 September 2026, https://digestai.news/story/xiaomi-releases-mimo-v2-6-pro-rl-with-1m-token-context-and-1-02t-param","publisher":"Digest AI","title":"Xiaomi releases MiMo-V2.6-Pro-RL with 1M-token context and 1.02T parameters","datePublished":"2026-09-21T20:06:48Z","url":"https://digestai.news/story/xiaomi-releases-mimo-v2-6-pro-rl-with-1m-token-context-and-1-02t-param"},"generatedBy":"Written by Digest AI's editorial model from the linked sources; the sources are the record.","license":"Headlines, digests and key points are written by Digest AI and may be quoted with a link to the story page. Linked articles belong to their publishers. Terms: https://digestai.news/terms#reuse"}