Anthropic reportedly released Claude Fable 5.2 and Mythos 5.2

AT A GLANCE

The report states that Anthropic released Fable 5.2 and Mythos 5.2, focusing on long-term complex technical projects, but official materials and independent verification are still lacking.

Article

Anthropic releases the strongest models before IPO: Fable and Mythos 5.2: A thousand AI collaborate to 8 create a system, formalizing Fermat's Last Theorem in 30 hours

Breaking news from Beating AI: Anthropic has released Claude Fable 5.2 and Mythos 5.2. These two models can now handle engineering and research projects that previously required professional teams to complete over long periods, including developing operating systems from scratch, formally proving mathematical theorems, and solving long-standing unresolved problems in black hole theory.

Anthropic allowed Fable 5.2 to begin developing an operating system starting from an empty folder. It coordinated over 1000 sub-agents, working continuously for 8 days, autonomously dividing tasks, programming, testing, and reviewing code. The resulting system could boot on a laptop, featuring an independent kernel, disk and network drivers, a file system, graphical windows, and an installer.

In another experiment, approximately 1000 Fable 5.2 agents spent 30 hours completing the formalization of an existing proof of Fermat's Last Theorem—transforming the mathematical proof into code that a computer can verify step by step. Previously, Anthropic had required specialized tools and human researchers, taking 11 days to finish similar tasks. Under the same configuration, Opus 5.5 ran for about 30 hours but only completed 26%。

Fable 5.2 and also helped solve two unsolved mysteries in black hole theory. One of these involved complex calculations proposed in 2003 years ago that had previously been difficult to complete.

Anthropic also enabled it to independently design three electronic devices, including circuit boards, components, housings, and software. After engineers from partner companies manufactured and assembled them according to the designs, all three devices functioned properly without any circuit modifications.

In official evaluations, Fable 5.2 achieved 2.0 on FrontierCode 49.1% programming tests, surpassing Opus 5.5's 45.5%. Robot simulation tests improved from the previous generation's 27.6% to 64.9%。Mythos 5.2, targeting cybersecurity and biological research; in known vulnerability rediscovery tests, it reached 89.6%, compared to the previous generation's 78.3%.

Fable and Mythos use the same set of model weights, with the main difference being security restrictions. Fable is open to all users, while Mythos relaxes restrictions on cybersecurity and biological research and is currently available only to vetted institutions.

Anthropic also proactively limited the model's training progress on tasks such as AI chip kernel optimization and model architecture improvements, keeping this capability roughly at the level of Opus 5.5 to reduce the risk of accelerating development of stronger AI versions of AI.

Key points

According to the report, Anthropic released Claude Fable 5.2 and Mythos 5.2.
According to the report, Fable 5.2 scheduled more than 1000 sub-agents to work continuously for 8 days, developing an operating system that can be booted on a laptop starting from an empty folder.
According to the report, about 1000 Fable 5.2 agents completed the formalization of an existing proof of Fermat's Last Theorem in 30 hours; under the same configuration, Opus 5.5 took about 30 hours to complete 26%.
The report states that the three electronic devices designed by Fable 5.2 all function properly after being manufactured and assembled by engineers at partner companies, without requiring any circuit modifications.
The article states that Fable 5.2 achieved 49.1% in the FrontierCode 2.0 test, while Opus 5.5 scored 45.5%; robot simulation test scores rose from 27.6% to 64.9%.
The article states that Mythos 5.2 achieved 89.6% in the known vulnerability rediscovery test, compared to 78.3% for the previous generation.
The report states that Fable and Mythos use the same set of model weights; Fable is available to all users, while Mythos is provided only to vetted institutions.
The API prices given in the text are 8 USD per million input Tokens and 40 USD per million output Tokens, and it is stated that the prices are 20% lower than the previous generation, while the generation speed is improved by up to 30%.
According to the report, Anthropic recommends using Opus 5.5 for general programming and everyday knowledge tasks, while Fable and Mythos are mainly intended for complex, long-running technical projects.

AI-assisted interpretation

The following is analysis, separate from reported facts. Verify important claims independently.

The core claim of this news brief is: Anthropic has launched two models better suited for long, multi-step technical tasks. Fable 5.2 leans toward complex engineering and research tasks, while Mythos 5.2 relaxes restrictions on cybersecurity and biological research, and is therefore only available to vetted institutions. The article lists operating system development, formalization of mathematical proofs, electronic device design, and multiple benchmark scores, but these are currently all reported in this news brief.

Why it matters to readers

If the capabilities, pricing, and open-access arrangements reported are confirmed, such models could enable more large-scale engineering and research tasks to be completed through collaboration among multiple AI agents, and could also affect performance and pricing competition in model services. The article also notes that Anthropic has restricted training progress for its models on certain AI research tasks, reflecting the trade-off between capability enhancement and safety risks.

Beginners should focus on understanding two points: first, “multi-agent collaboration” refers to many AI sub-agents dividing the work to complete the same project; second, the evaluation percentage is only a result from a specific test and cannot be directly equated with everyday user experience. For ordinary tasks, the text still recommends prioritizing Opus 5.5.

Risks and unknowns

The material comes from second-hand news on media channels and does not include an official Anthropic announcement, model documentation, or API pricing page.

Anthropic reportedly released Claude Fable 5.2 and Mythos 5.2

Article

AI-assisted interpretation

Why it matters to readers

Risks and unknowns

Related Developments

Related concepts

Keep following

Security Center