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
- antirez backs low-bit DeepSeek 4.1 Flash over Qwen 27B dense on Apple M5 Macs — antirez · 2026-11-12
- Gemini re-renders images as text when vision is removed, dropping only 6% WSR — zacharynado · 2026-11-12
- Bare-harness coding benchmark: Fable 5.2 best overall, Gemini 4 Carbon best value — Semantic_meaning · 2026-11-12
- Google's Physical-Engineering's Last Exam: all 14 frontier models score below 50% — dl_weekly · 2026-11-12
- Reddit's 2026 LLM hall of fame: Opus 5.5 still unbeaten after 9 rival releases — G0dZylla · 2026-11-12
- Commit leak reveals unreleased Gemini 4 variants: Argon, Hydrogen and Helix (unconfirmed) — BlackHC · 2026-11-11