AnalysisAI ModelsSeptember 20, 2026

Analysis claims 39% of Claude Code calls get zero thinking tokens

A 65-day analysis of 43,000+ Claude Code invocations found 39% of Fable 5 calls receive zero thinking tokens, with a median of just 123, while benchmarks use 16K-128K. The poster argues Anthropic is quietly cutting reasoning budget while presenting the same model.

1 source

More stories today

Open the live feed