2026 年 9 月模型榜前五：Astra 以 84.08 分居次
榜单显示 Claude Fable 5.1 第一、GPT-6 Astra 第二，并指出前五里已有两个被自家新版本取代。
The top 5 AI models right now, ranked by real benchmark score. LLM leaderboard, September 2026 - the top 5, broken down > Claude Fable 5.1 (Anthropic) - score 84.61, #1. 1M context, $10/$50 per M tokens, 66 tok/s, 304s latency. current flagship, closed license. > GPT-6 Astra (OpenAI) - score 84.08, right behind it. 1.05M context, same $10/$50 pricing, 54 tok/s, 340s latency. estimated benchmark, still current. > Claude Opus 5 (Anthropic) - score 81.97. cheaper at $5/$25, 52 tok/s, fastest of the top 3 at 108s latency. > Claude Fable 5 (Anthropic) - score 81.43. already SUPERSEDED by 5.1 above it. same $10/$50 pricing, slightly larger 1M+ context. > GPT-5.6 Sol (OpenAI) - score 80.63. also superseded. cheapest of the five at $4/$20, 60 tok/s, 164s latency. two of the top 5 are already outdated versions of the ones above them how fast is "state of the art" actually moving right now?