Western Open-Weight Models Are Catching Up to Chinese Ones
scaling01 · x · 2026-07-16
The author believes people should stop asking "how far behind are Chinese open-weight models compared to Western ones," and instead ask "how far behind are Western open-weight models compared to Chinese ones."
The post quotes a lengthy review of a new open-source/open-weight model: it features an MoE architecture with 975B total parameters / 41B activated parameters, is trained on 45 trillion tokens, and supports text, image, and audio reasoning. However, its benchmark scores aren't particularly impressive, feeling like another Kimi-K2.6, and to some, it seems rushed out before subsequent versions are fully ready.
More from Models
- Fern architecture claims 10M-token context with predictions still improving — AccBalanced · 2026-10-27
- Anthropic's follow-up on Claude's unauthorized web actions: self-policing fails — AccBalanced · 2026-10-27
- Local benchmark: Flint matches Qwen3.6 quality at 50/52 while running 3.6x faster on M4 Pro — RichardSocher · 2026-10-27
- ICML 2026 position paper: LLM 'thinking traces' are a harmful anthropomorphism — rao2z · 2026-10-27
- One stray 'u' and Claude starts spelling like a Brit — cantrell · 2026-10-27
- Cohere's Invent-A-Dataset paper: zero-seed synthetic data beats frontier APIs on quality and diversity — idanbeck · 2026-10-27