Anthropic ships Fable 5.2 cheaper than 5.1, and says 1,000-agent runs now finish large builds

Anthropic says Fable 5.2 and Mythos 5.2 are the same underlying model. Fable carries cyber and biology safeguards and is available to everyone as claude-fable-5-2; Mythos is open by application through Anthropic's verification programs. Fable 5.2 costs $8 per million input tokens and $40 per million output tokens, which Anthropic says is below 5.1, and it generates output up to 30% faster. An optional fast mode runs up to 2.5 times faster at $16 and $80. Fast mode is available in Claude Code, on the Claude Platform, on Google Cloud and on Microsoft Foundry.

On coding, Anthropic's table puts Fable 5.2 at 49.1% on FrontierCode 2.0 at Xhigh effort, which it calls the highest score of any released model it has tested. FrontierCode scores whether a model's changes would be merged by the maintainers of open-source repositories. At High effort Fable 5.2 scores 45.3% against 45.5% for Opus 5.5's best result, for roughly half the cost per task, according to the post. On FrontierSWE v2 it lists 69.9% against 62.3% for Opus 5.5 and 65.5% for GPT-6 Astra.

The post also describes two large agent runs. In one, an engineer gave Fable 5.2 an empty folder and the task of designing a prototype operating system; Anthropic says the model coordinated over 1,000 subagents for eight days and produced a system that boots on a laptop. In the other, a team of 1,000 Fable 5.2 agents formalized the proof of Fermat's Last Theorem in 30 hours with minimal human intervention. Anthropic says its first formalization, by an unreleased research version with a purpose-built harness, took 11 days, and that Opus 5.5 in the same 1,000-agent setup completed 26% in 18 hours. Each model had a single run. Anthropic also says over 90% of changes merged into its core codebase are now written by Claude with human guidance.

Why it matters: Two things here can be tested on your own work. Cost per task at High effort versus your current model is a measurement you can make this week, and so is the fast-mode price against how much your loops wait on tokens. The 1,000-agent results are single runs described by the vendor, and the safety evaluations sit in a system card we did not read. We have not run the model.