GPT-6 Astra and Claude Fable 5.1: performance scores versus task costs
Arena says GPT-6 Astra (Max) and Claude Fable 5.1 (Max) cost far more per task than other models from their respective labs, despite relatively close performance.
TLDR
Arena’s September 16, 2026 comparison weighs net improvement scores against median cost per task for the top model in each family, defined as variants sharing the same model base name.
For OpenAI, it lists GPT-6 Astra (Max) at +11.7% and $3.94 per task, versus GPT-5.6 Sol (xHigh) at +7.0% and $1.03.
For Anthropic, it lists Claude Fable 5.1 (Max) at +13.7% and $4.40 per task, versus Claude Opus 5 (High) at +10.2% and $2.07.
Arena characterizes Astra and Fable as costing far more than other models from their respective labs despite relatively close performance.
Combined views
4.7K
1 Source, first seen 14d ago