Anthropic maps Claude use: Fable 5.2 for hard tasks, Opus 5.5 default

The guide arrives alongside Fable 5.2, which Anthropic described as its top model and which it says improves coding and visual-information processing over the previous model, with output speed up to 30％ faster. Rather than positioning the newest model as the default, Anthropic's advice keeps Opus 5.5 as the starting point for many tasks at medium effort, and reserves Fable 5.2 for advanced problems in coding, finance and science.

Anthropic also frames effort as a lever to try before switching models, with a caveat: settings above "high" can carry costs comparable to a higher-tier model, so it recommends comparing a lower-effort top model. Beyond model choice, the post points to orchestration tricks — assigning lower-tier models to subagents that handle specific parts of a task, and using an advisor tool to consult a higher model on hard problems mid-task.

For teams budgeting Claude usage, the practical takeaway appears to be that model tier and effort level should be tuned together, not treated as separate switches. The post does not give per-task pricing figures, so the cost comparison it recommends would have to be run in-house.