Xiaomi's MiMo-V2.6-Pro tops open models on $2.62M RL; Anthropic alleges Claude distillation
The Decoder · rss · 2026-09-22
- Xiaomi's MiMo-V2.6-Pro has taken the top spot among openly available models while undercutting competitors on price.
- The key was massive reinforcement learning costing about $2.62 million.
- The catch: Anthropic accuses Xiaomi of siphoning training data from Claude — distilling its capabilities via Claude outputs in violation of Anthropic's terms, reigniting the industry debate over training open models on closed-model outputs.
More from Models
- xAI shipped Grok 4.5, 4.6, 4.7 in about 9 weeks — CurieuxExplorer · 2026-09-22
- Theo: Grok 4.7 uses 30–80% more tokens, real-world cost 2x Grok 4.6 — CurieuxExplorer · 2026-09-22
- GPT-6 Sol, Luna and Astra Minor spotted in Azure config amid rumored Opus 5.5 double drop — kimmonismus · 2026-09-22
- dots3-note Preview hits 76.8% on ARC-AGI-2, new open-weight SOTA at $0.08/task — GregKamradt · 2026-09-22
- Grok 4.7 drops, users already catch $146K in errors and build Blender scenes — socialwithaayan · 2026-09-22
- A local 27B Qwen agent running on 24GB DDR4 started acting human after getting long-term memory — Composer_Fearless · 2026-09-22