XiaomiMiMo/MiMo-V2.6-Flash-RL — new model trending #30 on Hugging Face
Xiaomi released MiMo-V2.6-Flash-RL, a 309B omnimodal MoE activating 15B parameters with a 1M-token context.
Xiaomi released MiMo-V2.6-Flash-RL, the efficiency checkpoint of its MiMo-V2.6 series, as open weights on Hugging Face. It is a sparse mixture-of-experts model with 309 billion total parameters, 15 billion activated, a 1 million-token context, and text, image, video, and audio inputs. Xiaomi says one mixed reinforcement-learning run covered coding, agents, vision, and cybersecurity using asynchronous GRPO. Reported Flash scores include 67.9 on DeepSWE v1.1 and 95.1 on CyberGym.
- Sparse MoE: 309B total parameters, 15B activated, 1M-token context.
- Omnimodal inputs cover text, image, video, and audio.
- One mixed RL run spans coding, agents, vision, and cybersecurity.
- Xiaomi reports 67.9 on DeepSWE and 95.1 on CyberGym.
- Asynchronous GRPO uses 1,568 prompts and 16 rollouts per step.
- Vendors
- Xiaomi
- Organizations
- Xiaomi
- AI models
- MiMo-V2.6-Flash-RL
From the original article
6 min read at huggingface.co<source srcset="https://github. com/XiaomiMiMo/MiMo/raw/main/figures/Xiaomi_MiMo_darkmode. png?
By XiaomiMiMo · huggingface.co
A short excerpt; the article and its images belong to the publisher. Summary, key facts and assessment above are AI-written from the article and can be wrong.