Anthropic says to start with Opus 5.5 at medium effort, reserves Fable 5.2 for hardest tasks

Anthropic recommends that most users begin with Opus 5.5 at a "medium" effort setting when choosing a Claude model. The company laid out the guidance in an official blog post on November 9 (US time), according to Japanese outlet ITmedia AI+ (the source is in Japanese). The post covers how to split work across Fable 5.2, Haiku 5.5, Sonnet 5.5 and Opus 5.5 in Claude Code and Claude Platform.

Which model for which job

Anthropic's recommendations, as reported by ITmedia:

• Opus 5.5: the default starting point for many tasks, tried at medium effort. Effort is a setting that adjusts how many tokens the model spends on output.
• Sonnet 5.5: work with a clearly defined scope, such as bug fixes.
• Haiku 5.5: simple processing run at high volume, such as classification and information extraction.
• Fable 5.2: advanced problems in coding, finance and science.

Adjust effort before switching models

Anthropic says changing the effort setting is worth trying before changing models. It cautions that settings above "high" can cost about the same as using a higher-tier model. The company suggests comparing costs by running a higher-tier model at reduced effort, and validating per task.

The post also describes two other approaches:

• Assigning a lower-tier model to a subagent that handles a specific part of a larger task.
• Using an advisor tool, which lets the working model ask a higher-tier model for advice on difficult parts of a task.

Fable 5.2

ITmedia describes Fable 5.2 as Anthropic's top-tier model, announced the same day as the guidance. Compared with its predecessor, Anthropic says it improves coding and visual information processing. The company also claims output speed is up to 30% faster. It is available from the day of announcement on paid Claude plans and through the API.

Context window size and per-token pricing for Fable 5.2 are not yet disclosed here, and no benchmark scores have been published in this report.

What this means

The guidance treats model choice as a cost-and-capability decision with several levers, not a single selection. Anthropic's default is Opus 5.5 at medium effort, with Fable 5.2 positioned for the hardest coding, finance and science problems. Its caution that effort settings above "high" can approach the cost of a higher-tier model means users should measure spend at each effort level before moving up a tier. Subagents on lower-tier models and the advisor tool give teams a way to use a top-tier model on only the hard parts of a task. Anthropic's speed figure for Fable 5.2 is a company claim and should be checked against real workloads.