Anthropic releases Claude cost-saving guide: start with lower intensity, escalate upon failure, saving 41% while improving accuracy.

Beating AI News Flash: Anthropic has released a Claude model selection guide, advising developers not to default to the most powerful Fable 5.2. Most programming and knowledge work should start with Opus 5.5 at Medium reasoning effort. For feature development with clear requirements and routine automation, the cheaper Sonnet 5.5 can be used; simple tasks such as classification and information extraction should be handled by Haiku 5.5. For large-scale code migration, complex scientific research, or tasks that Opus struggles to complete, consider Fable 5.2.

Anthropic believes that one effective way to reduce costs is to let AI check its own results. For example, when programming, run automated tests and fix errors if found, rather than starting with the highest reasoning effort.

On the same subset of the SWE-bench Pro programming test, the official team had Opus 5.5 first execute with Low reasoning effort and only retry failed tasks with High. The result was a pass rate of about 97%, with an average cost of $0.17 per question; using High directly for all tasks yielded a pass rate of 95.3%, with a cost of $0.29 per question. The former reduced costs by about 41% while achieving a slightly higher pass rate.

Simply increasing reasoning effort also has a cost-effectiveness ceiling. When Opus 5.5 was upgraded from High to Xhigh, its score on SWE-bench Pro improved by only about 1.4 percentage points, while costs reached 2.5 times the original. Anthropic suggests that if a model still cannot solve a difficult problem after reaching High, one can try a stronger model at a lower reasoning effort.

For example, in the FrontierCode 2.0 programming evaluation, Fable 5.2 using High achieved 45.3%, nearly matching Opus 5.5 using Max at 45.5%, while the average cost per question was only about half of the latter. Although Fable's API token unit price is twice that of Opus, in such highly difficult tasks, the total cost of completing the task may actually be lower.