Pick

Anthropic Appears to Be A/B Testing Reduced Reasoning Effort in Claude Code

·2026.08.22 09:00

Key point

Anthropic is reportedly conducting server-side A/B tests to lower reasoning effort levels in the latest versions of Claude Code.

Details

It has been confirmed that Anthropic is conducting server-side experiments to lower the reasoning effort scale in some 'Fable 5' sessions of Claude Code 2.1.236 or later.

As a result, the model's reasoning depth and response quality may vary between users even for the same prompt.

Applicable to: Claude Code versions 2.1.236 or later and Fable 5 model sessions
Excluded: Previous versions of Claude Code and the Opus 5 model
Experiment method: Estimated to be a server-side A/B test where experiment enrollment conditions are set based on the client version

reasoning anthropic claude-code ab-testing llm-infrastructure

Read originalShare on Threads

This summary was generated automatically by AI. Check the original for the author's claims and context. Copyright belongs to the original author.

Our guide explains how the AI works. Report summary errors, attribution issues, or removal requests via Contact.