Anthropic Launches Claude Fable 5.2 and Mythos 5.2 With Faster Output and Lower Prices

Anthropic has released Claude Fable 5.2 and Claude Mythos 5.2, its latest models for demanding coding and research work. The company claims gains in reasoning and vision, with output up to 30% faster than the 5.1 models and lower API prices.

The two names refer to the same underlying model. Fable includes stronger cybersecurity and biology restrictions for general availability, while Mythos is offered to vetted organizations through verification programs. For everyday coding and knowledge work, the company still recommends starting with Claude Opus 5.5 and moving up when a task needs more capability.

Standard API pricing is $8 per million input tokens and $40 per million output tokens, down 20% from the 5.1 models' $10 and $50 rates. An optional fast mode promises up to 2.5 times faster token generation for $16 and $80 per million input and output tokens, respectively.

Benchmark Results

Anthropic's published results put Fable 5.2 ahead of Opus 5.5 across most of the tests below, though not every measure. Gemini 4 Carbon scores higher on AutomationBench, and Opus 5.5 retains a higher AA-Briefcase Elo rating. These are the company's reported evaluations, with the model-specific qualifications preserved:

Scores use the highest available effort unless noted. Fable and Opus results include safeguards unless otherwise specified; the cybersecurity and protein-design entries marked Mythos are not Fable scores. On CyScenarioBench, each lab uses unique tasks, so those results are not directly equivalent. RoboDojo Sim results come from Anthropic's internal runs, and the OSWorld result covers a partial offline set.

Long-Running Coding and Engineering Work

In one experiment, an engineer asked Fable 5.2 to build a prototype operating system from an empty folder. The company says it coordinated more than 1,000 agents over eight days, producing a system that boots on a laptop with storage, drivers, windows, a login screen, and an installer. Another agent team migrated a production service from Python to Rust, with startup about 26 times faster.

The hardware tests went beyond software. Working with circuit-board manufacturer Diode, Fable 5.2 designed a home voice-assistant hub, a USB speakerphone, and a handheld virtual pet, including their circuit boards, enclosures, and software. Diode manufactured the boards, and Claude guided engineers through testing and assembly; all three worked without electrical modifications, according to Anthropic.

Other prerelease examples included work on two open problems in black-hole physics and a 1,000-agent effort to formalize the existing proof of Fermat's Last Theorem in 30 hours. That mathematical task involved translating the proof into a computer-checkable form, not discovering the theorem or its original proof.

Availability and Safeguards

Fable 5.2 is available across Claude's platforms and cloud partners including AWS, Google Cloud, and Microsoft Azure, with the API identifier claude-fable-5-2. Mythos access remains subject to the Cyber or Life Sciences Verification Programs; enterprise customers can also use it through Claude Security.

Fable still restricts high-risk cybersecurity work and much research biology. The updated safeguards produced about 55% fewer blocks in early testing than Fable 5.1, but that does not remove the access restrictions. Eligible enterprise customers can use zero data retention with the Enterprise Frontier Safeguards system, now in private beta; other customers have 30-day retention for safety monitoring.

For launch week, eligible plans and seats that include Fable receive one to three usage resets based on subscription tenure. Fable is also included for up to 50% of weekly usage limits on Pro plans and standard Team seats.