最新模型
Anthropic Releases Claude Fable 5.2, Cutting Prices and Tightening Reasoning-Block Conversation Binding
Claude Fable 5.2 and Mythos 5.2 launched on November 9, with standard input and output prices 20% lower than their predecessors. The update also broadens prefix checks for reasoning blocks, so agent frameworks that edit message history or switch models should check compatibility.
Anthropic released Claude Fable 5.2 and Mythos 5.2 on November 9, targeting long-running agent tasks, software development, and complex reasoning. The two models use the same underlying model, but have different access and safety restrictions: Fable is available to general users, while Mythos is limited to verified organizations. Anthropic says standard-mode output speed is up to 30% faster than the previous generation. A paid fast mode offers up to 2.5 times the output speed. Official announcement
The API documentation lists a 1 million-token context window and a maximum output of 128,000 tokens. Standard input and output cost $8 and $40 per million tokens, respectively, 20% less than the $10 and $50 prices for 5.1. Adaptive thinking remains enabled, and the API defaults to high reasoning effort; developers must use effort to control computational investment. Model documentation
On performance, Anthropic reports that Fable 5.2 scored 49.1% on FrontierCode 2.0 at the Xhigh setting, compared with 32.4% for its predecessor. At the High setting, it scored 45.3%, close to Opus 5.5’s top score of 45.5%, at about half the cost per task. These are vendor-reported benchmark results and should be validated against an organization’s own codebase, tool permissions, and retry strategy; they should not be treated as the success rate after deployment. Benchmark and configuration details
The reasoning-block binding change is especially important for migration. The new version checks the system, tools, and message history preceding a block for all accounts. If that content has been modified but the original reasoning block is still returned, the default behavior is to return HTTP 400. As a result, workflows that rewrite history and worked on Fable 5.1 for some legacy accounts may fail after upgrading. Anthropic offers a beta option to discard invalid blocks, but warns that this may increase output tokens. Migration changes
Cross-model routing also has a directional limitation: only Fable 5.2 and Mythos 5.2 can read reasoning blocks generated by the new versions. The legacy computer-use tool computer_20251124 returns an error directly. Engineering teams should test history compaction, tool-list updates, and model switching before comparing task costs at different effort levels. In addition, both models generally require 30-day data retention; zero data retention requires special authorization, which may affect adoption conditions for some enterprise workloads. Compatibility and data-retention requirements