llama.cpp can now convert MiMo-V2.6 Pro and Flash, one day after Xiaomi open-sourced them
GitHub·medium signal
ggml-org/llama.cpp PR #29257, merged 2026-09-22, adds conversion for Xiaomi's MiMo-V2.6 Pro and Flash. It moves the Kimi-K3 mxfp4 expert repack into base.py so DeepSeek-V4, Kimi-K3 and MiMo share it. The author reports both models load with vision and needed no runtime changes, but the audio decoder in the mmproj is excluded. You can run MiMo-V2.6 locally in GGUF now, without speech output.