Anthropic ships Claude Fable 5.2 at a 20 percent lower price and keeps Mythos 5.2 behind vetted access

Anthropic released Claude Fable 5.2 and Claude Mythos 5.2 on November 9. It calls them its most capable models for coding, finance and the sciences. The split follows the pattern Anthropic set with the version 5 generation. Fable is the general release with safeguards. Mythos is a restricted-access version for organizations that pass Anthropic's cyber and life-sciences verification programs, according to prior coverage of the Mythos line.

The most immediate commercial change is price. Several trackers report Fable 5.2 at $8 per million input tokens and $40 per million output tokens, down from $10 and $50 for Fable 5.1. They also report a fast mode at $16 and $80, a 1M-token context window and 128K maximum output.

On capability, every number available so far comes from Anthropic. Mythos 5.2 scores 89.6 percent on CVE Rediscovery, against 78.3 percent for Mythos 5.1, and 76.3 percent on CyScenarioBench, against 61.7 percent. On a 24-hour de novo protein-binder design task it scores 85.1 percent, against 79.1 percent. Fable 5.2 reportedly reaches 49.1 percent on FrontierCode 2.0 at its highest effort setting, 69.9 percent on FrontierSWE v2 and 74.2 percent on Humanity's Last Exam with tools. Outlets disagree on the Terminal-Bench-Science 0.1 figure, and cached copies of Anthropic's table leave several rows blank. The comparison set is Anthropic's own earlier models (Mythos 5.1 and Opus 5.5). There is no head-to-head against OpenAI's or Google's current models on the same harness.

The most striking claim is mathematical. The Russian-language channel AI ML Big Data summarized Anthropic's post as saying that 1,000 agents formalized the proof of Fermat's Last Theorem in 30 hours. Anthropic's September formalization took dozens of agents 11 days, produced about 13 million lines of Lean (a computer proof-checking language) and was checked against Mathlib, Lean's library of formalized mathematics, including its statement of the theorem. The 30-hour result has no published proof artifact yet. It measures wall-clock throughput from parallel agents on a target already solved once. It does not show new mathematical ability.

The release comes days before Anthropic's expected Nasdaq debut. A visible capability improvement and a price cut in the same week strengthen the company's pitch to investors. That timing does not make the numbers inaccurate, but it is a reason to wait for independent verification.