anthropic/claude-fable-5.2

Anthropic's model for the most demanding reasoning and long-horizon agentic work, at 20% less per token than Fable 5

23 runs
Playground API Examples README

Readme

Claude Fable 5.2

Claude Fable 5.2 is Anthropic’s model for the most demanding reasoning and long-horizon agentic work. It improves on Fable 5.1 in coding, reasoning, and vision, and Anthropic says it generates output up to 30% faster. It costs 20% less per token than Claude Fable 5 on Replicate.

Anthropic recommends starting most work on Claude Opus 5.5 and moving up to Fable 5.2 when a task is too hard or too open-ended for it.

What it’s good at

  • Coding. Long, ambiguous engineering work such as large migrations, hard bugs, and multi-file changes. On FrontierCode 2.0, Anthropic reports that Fable 5.2 at high effort matches Opus 5.5’s best score for about half the cost per task.
  • Research and analysis. Data-heavy questions, financial models, legal review, and research synthesis, where judgment about ambiguous evidence matters.
  • Science and math. Multistep technical reasoning in physics, biology, and mathematics.
  • Vision. Reading charts, diagrams, screenshots, and documents sent alongside your prompt.

Thinking effort

Fable 5.2 always thinks before it answers. You can’t turn thinking off, but you can control how much it does with the effort input:

  • high (default) is Anthropic’s recommended starting point and gets most of the model’s capability.
  • low and medium are faster and cheaper, and work well when your task doesn’t need deep reasoning.
  • xhigh is for the hardest long-running coding and agentic tasks.
  • max removes limits on token spending. It is slower and costs more, often for a small gain over xhigh.

Thinking tokens count toward max_tokens and are billed as output tokens. The default max_tokens is 32,000. At xhigh or max, set it to 64,000 or more so the model has room to think and answer.

Inputs

  • prompt: your input prompt.
  • system_prompt: an optional system prompt to steer the model’s behavior.
  • image: an optional image to send with your prompt. Images are priced as (width px × height px) / 750 input tokens.
  • max_tokens: the maximum number of output tokens, including thinking. Up to 128,000.
  • effort: how much thinking the model does before responding.
  • max_image_resolution: scales the image down before sending it, to save time and money.

Specs

  • Context window: 1 million tokens
  • Maximum output: 128,000 tokens
  • Knowledge cutoff: June 2026

Pricing

Fable 5.2 is billed per token: $8 per million input tokens and $40 per million output tokens.

Safeguards

Fable 5.2 has safety classifiers for high-risk cybersecurity and biology requests. Anthropic answers a blocked request with a different Claude model instead of Fable 5.2. If Anthropic’s safety systems decline a request outright, the run fails with a content error instead of returning an empty response.

Model created
Model updated