Claude Fable 5.1 was announced with claims of setting new standards for coding, knowledge work, and problem-solving tasks. It demonstrates notable improvements on the Terminal-Bench-Science 0.1 benchmark, scoring 52.6%, up from previous versions.
The model offers five reasoning levels: low, medium, high, xhigh, and max, with no option to disable reasoning. Tests showed that at low and medium levels, reasoning traces were often skipped, resulting in similar outputs and costs. Higher levels, especially xhigh and max, produced significantly longer outputs with detailed reasoning traces.
At xhigh, output tokens increased dramatically, with reasoning details including specific design choices for generating a pelican SVG. The max setting produced the most detailed reasoning, with the longest output and highest cost, indicating deeper problem-solving capabilities.
These results matter for engineers running models in production, as they highlight how reasoning effort levels impact output length, cost, and task performance. The benchmark scores and reasoning behaviors inform model configuration decisions.
Source: https://simonwillison.net/2026/Sep/1/claude-fable-5-1/