Claude Fable 5.2 drops to $8/$40 and runs up to 30% faster

Anthropic on November 9 opened Claude Fable 5.2 on all paid plans, the Claude Platform and the major clouds at $8 per million input tokens and $40 per million output tokens, down from $10 and $50, with output up to 30% faster than Fable 5.1. Anthropic’s model documentation also lists request shapes the model rejects, so API code that forces tool use gets a 400 error, and the docs say to drop temperature, top_p and top_k from requests.

What changes in your requests

Adaptive thinking is always on and can’t be turned off. Depth is set with the effort parameter, which defaults to high on the API. In the Claude apps and Claude Code the default is Medium. Fable 5.1 defaulted to High in Claude Code, Anthropic said in September, so Claude Code now starts Fable one effort level lower than it did on 5.1.

Forced tool use, the tool_choice types any and tool, returns a 400. The docs say to omit temperature, top_p and top_k, and point to a separate page for the sampling values that return a 400 error.

Editing earlier turns of a conversation is the change most likely to surface in agent code. Anthropic says that to help prevent distillation, where another model is trained on Claude’s outputs, preserved thinking now covers every API account on Fable 5.2. Editing prior context invalidates the thinking blocks and can return a 400. Anthropic’s suggested routes are compaction, context editing or mid-conversation system messages, and Claude Code, Managed Agents and the Agent SDK handle this on their own.

Fable 5.2 also requires 30-day data retention and isn’t available under zero data retention unless Anthropic expressly authorizes it. In September, eligible customers could run Fable 5.1 with zero data retention until Enterprise Frontier Safeguards, which keep data in cloud infrastructure the customer controls, reached them in phases. A team that relied on that arrangement for 5.1 needs Anthropic’s authorization to keep it on 5.2.

The docs also cover classifier refusals, with a retry on another Claude model, and a faster output mode for Messages API requests at premium pricing.

Fable 5.2 or Opus 5.5?

Opus 5.5 costs half as much, $4/$20, with the same 1M-token context, 128K max output and June 2026 knowledge cutoff, and defaults to medium effort. Anthropic’s own early FrontierCode 2.0 run put Fable 5.2 at High effort about level with Opus 5.5 at Max effort for roughly half the cost per task; at Xhigh, Fable 5.2 scored the highest of any released model tested. Anthropic still recommends starting most coding and knowledge work on Opus 5.5, which it described in September as matching Fable 5.1 on most work, and keeps Fable and Mythos for the most ambitious, open-ended technical work.

Artificial Analysis gives an independent view of the top setting. Fable 5.2 at Max scores 65 on its Intelligence Index, the highest so far, against a median of 40 for comparable models. It took 45.0 minutes per index task against a median of 12.9, cost $26.54 per task and generated 275M tokens over the run, against a median of 72.8M.

What Anthropic says it can do

The capability claims are Anthropic’s own. It says Fable 5.2 and Mythos 5.2 sit a step above Opus 5.5 in almost all capabilities it evaluated and performed as well as or better than its recent models on most alignment measures. External evaluators, including the UK AI Security Institute, Frontier Design and METR, tested the model before release.

A team of 1,000 Fable 5.2 agents formalized Fermat’s Last Theorem in 30 hours, turning the proof into a language a computer can check step by step; with the same setup, Opus 5.5 completed 26% in 18 hours and barely progressed after 30. Mythos 5.2 accurately modeled 28 of the 34 gene-therapy capsids it attempted. In prerelease testing, Anthropic says, the models “designed three working electronic devices from scratch” and “solved two open problems in the physics of black holes”. On RoboDojo, a simulated two-arm robot benchmark, Fable 5.2 outperformed other models and most AI systems built to control robots.

Plans and Mythos access

Ultra and Team Premium subscribers get Fable limit resets by subscription length: one under six months, two for six to 12 months, three beyond 12, usable until March 8, 2027. Pro and Team Standard get a week to try Fable 5.2, using up to half the weekly limit, plus one reset.

Mythos 5.2 is the same underlying model with the same specifications and pricing; Fable carries strong safeguards against misuse. Mythos goes only to organizations verified through Anthropic’s programs, such as the Life Sciences and Cyber Verification Programs. To get it, apply to the program that covers your use case or contact your account team.