State of AI · 2026-10-16
The AI race at a glance
Editorial picks by AIRaceBoard based on capability tier and price; not a benchmark score.
🏆 #1 overallClaude Fable 5.1🇺🇸 Anthropic$10 / $50 per 1M
🇺🇸 #1 AmericanGPT-6 Astra🇺🇸 OpenAI$10 / $50 per 1M
🇨🇳 #1 ChineseKimi K3🇨🇳 Moonshot AI$3 / $15 per 1M
🔓 #1 open modelKimi K3🇨🇳 Moonshot AI$3 / $15 per 1M
💻 #1 for codingClaude Opus 5.5🇺🇸 Anthropic$4 / $20 per 1M
💸 Cheapest goodDeepSeek V4.1 Flash🇨🇳 DeepSeek$0.3 / $1.2 per 1M
US vs China vs Europe
🇺🇸 United States16 modelsTop pick: GPT-6 Astra
🇨🇳 China8 modelsTop pick: Kimi K3
🇪🇺 Europe1 modelsTop pick: Mistral Small 4
What changed this week
- Meta released Muse Spark Open (Oct 15), now the most capable US open-weight model per Meta: 816B total / 42B active, 1M context, $1.25/$4.25 per 1M on OpenRouter.
- NVIDIA Nemotron 3.5 Super VL (Oct 15): open multimodal model, 1M context, from $0.10/$0.40 per 1M on OpenRouter.
- DeepSeek API was down ~5 hours on Oct 15 (08:11-13:04 Beijing); DeepSeek is crediting every API account its Oct 8-14 spend.
- Ant Group open-sourced Ling 3.1 Flash weights (MIT, 560B / 25B active) on Oct 16; not on the board yet while API pricing is promo-only.
- Gemini 4 Argon GA (Oct 13): intro $2/$10 per 1M through Nov 30, then $4/$20 (Google API pricing).
- Anthropic list prices updated Oct 7: Opus 5.5 $4/$20, Sonnet 5.5 $2/$10, Haiku 5.5 $0.10/$0.50.