Zbigniew Majewski(@majewskizby):MiMo V2.6 Flash looks like the sweet spot for local inference. It delivers roughly 90-97% of Pro's practical coding and agentic performance while using just 15B active parameters vs 42B. On 4x DGX Spark, rough estimates are around 50-60 tok/s for Flash vs 20-25 tok/s for a heavily quantized Pro.

2026.09.21 21:29

MiMo V2.6 Flash looks like the sweet spot for local inference. It delivers roughly 90-97% of Pro's practical coding and agentic performance while using just 15B active parameters vs 42B. On 4x DGX Spark, rough estimates are around 50-60 tok/s for Flash vs 20-25 tok/s for a heavily quantized Pro.

Xiaomi MiMo@XiaomiMiMo

2026.09.21 20:51

Introducing Xiaomi MiMo-V2.6 — Pro & Flash. Frontier intelligence, all the modalities, built in public. 🔹 Two omnimodal models, advancing through scaled reinforcement learning 🔹 Pro performs on par with Claude Opus 5 and GPT-5.6 Sol across most agent benchmarks 🔹 Pro scores 46 on the Artificial Analysis Intelligence Index — the highest among open-source models 🔹 Stronger coding, computer use, 3D reasoning and creative capabilities 🔹 Open model weights, technical report, RL environments and training code Blog：