Claude Fable 5.2 completed the formalization of Fermat's Last Theorem in two days, and Anthropic released two flagship models at once

Core Viewpoint

Performance Evaluation of Claude Fable 5.2 and Mythos 5.2

Benchmark Test Results

The benchmark test results below show that Claude Fable 5.2 and Mythos 5.2 outperform Opus 5.5 in almost all capabilities assessed by Anthropic.

From a horizontal evaluation perspective, Fable 5.2 leads in multiple complex tasks. In programming, it scored 49.1% and 69.9% on FrontierCode 2.0 and FrontierSWE v2 respectively, surpassing GPT-6 Astra's 42.8% and 65.5%, and also leading Opus 5.5. In the robot operation evaluation RoboDojo Sim, Fable 5.2 achieved 64.9%, a 37.3 percentage point increase from the previous generation Fable 5.1's 27.6%, clearly higher than Opus 5.5's 39.4% and GPT-6 Astra's 29.8%.

Additionally, it scored 89.3% in the computer operation evaluation OSWorld 2.1 and reached 74.2% in the multidisciplinary reasoning evaluation Humanity's Last Exam (with tools), both ranking first among the models evaluated in the chart. However, in the knowledge work evaluation AA-Briefcase, Opus 5.5 slightly led with an Elo score of 1800, while Fable 5.2 scored 1785. Some high scores in scientific research and cybersecurity came from the concurrently released Mythos 5.2.

Claude Fable 5.2 completed the formalization of Fermat's Last Theorem in two days, and Anthropic released two flagship models at once

Reasoning Intensity and Efficiency

Under the Medium reasoning intensity setting, which is also the default setting for Claude applications and Claude Code, Fable 5.2 requires fewer tokens to complete challenging tasks compared to Fable 5.1.

In early tests of the FrontierCode 2.0 evaluation, Fable 5.2 achieved results comparable to Opus 5.5 using Max reasoning intensity while using High reasoning intensity, with the cost of individual tasks being about half of the latter. At Xhigh reasoning intensity, Fable 5.2 also outperformed all other released models participating in the tests.

Claude Fable 5.2 completed the formalization of Fermat's Last Theorem in two days, and Anthropic released two flagship models at once

Programming

Currently, over 90% of the merged code changes in Anthropic's core codebase have been completed by Claude under human guidance.

Anthropic has used Fable 5.2 to fix long-standing bugs in minutes, optimize the operational performance of various services, and complete important software projects that were previously difficult to initiate due to excessive time or cost.

Previously, a team of agents powered by an early version of Fable 5.2 migrated a critical production service from Python to Rust. After the migration, the service's operational efficiency and reliability improved, and its startup speed increased to about 26 times the original.

Tests have also found that Fable 5.2 possesses strong engineering judgment capabilities. Even when faced with somewhat vague requirements, it can provide high-quality results.

In one experiment, an Anthropic engineer tasked Fable 5.2 with designing a prototype operating system from scratch. The model started from an empty folder, coordinated over 1000 sub-agents, and continuously wrote, tested, and reviewed code for 8 days without interruption. Ultimately, it successfully developed an operating system that could boot and run on a laptop.

This system has its own kernel, can drive the entire machine; it includes disk, keyboard, and network drivers, as well as persistent file storage, a window system, a login interface, and an installer, and it also has built-in Claude functionality.

Early testers were also impressed with Fable 5.2's performance in applying programming capabilities to various different tasks.

Cybersecurity

Claude Mythos 5.2 is currently Anthropic's most powerful model in cybersecurity engineering capabilities. Compared to previous models, it can discover more security vulnerabilities and is better at reproducing and fixing these vulnerabilities. It is important to note that Claude Fable 5.2's security protection mechanisms will intercept most tasks related to cybersecurity.

A cybersecurity company found that one-sixth of the vulnerabilities identified and confirmed by Claude Mythos 5.2 were not discovered by all other leading models they tested.

In Anthropic's internal evaluation CVE Rediscovery, Claude Mythos 5.2 found 10.2% more vulnerabilities than Claude Opus 5.5 and 14.4% more than Claude Mythos 5.1.

Claude Mythos 5.2 also shows significant improvements in judgment accuracy and confidence calibration. One tester found that previous models tended to exaggerate the severity of about 1 in every 20 vulnerabilities discovered, while Claude Mythos 5.2 did not exhibit this issue.

Claude Fable 5.2 completed the formalization of Fermat's Last Theorem in two days, and Anthropic released two flagship models at once

Users can utilize Claude Mythos 5.2's cybersecurity capabilities through the cybersecurity validation program. Anthropic is also continuously expanding and improving this program.

Even without joining the cybersecurity validation program, Claude Enterprise clients can use Claude Security. This service leverages Anthropic's most powerful model to scan codebases for potential vulnerabilities, and it has now integrated Claude Mythos 5.2.

A recent update further enhanced Claude Security's capabilities. It can now continuously monitor code repositories, track code updates, and ensure that identified vulnerabilities are fixed; it can also analyze with more contextual information and integrate with other tools used by users via API and MCP.

Science and Engineering

Many important works in the fields of science, engineering, and mathematics increasingly rely on programming capabilities. Whether analyzing data, building molecular models, designing components, or controlling experimental instruments, software can accomplish these tasks.

Below, Anthropic introduces how Claude Fable 5.2 and Claude Mythos 5.2 help researchers solve challenges in these fields and advance related research.

Theoretical Physics

Claude Fable 5.2 solved two unresolved problems in the field of theoretical physics. Both problems are related to black holes and involve the unification of two fundamental theories of modern physics—relativity and quantum mechanics.

The two achievements made by Claude are based on the theory of Holographic Duality. The core idea of this theory is that certain universes containing gravity can be accurately described by a quantum theory that completely lacks gravity.

The first question dates back to 1983. At that time, physicists had already proven that as these universes continued to heat up, black holes would suddenly form, just like water suddenly boiling when heated to a certain point.

Claude Fable 5.2 completed the formalization of Fermat's Last Theorem in two days, and Anthropic released two flagship models at once

Address: https://hexagonmath.org/2611.00175v1

However, the methods used at that time were only applicable to sufficiently large universes, as relativity only applies at such scales. In 2003, researchers proposed a computational task to determine whether the same phenomenon would occur in very small universes. In these universes, only quantum theory could be used for description. However, the computation was too complex and time-consuming, and no one had been able to complete it.

Fable 5.2 essentially completed this computation independently and found that even in very small universes, this transition would also occur suddenly.

The second question involves a technique that physicists have been using for more than 15 years. This technique is used to exclude possible quantum gravity theories in simplified universe models. Fable 5.2 found the precise applicable boundaries of this technique.

Claude Fable 5.2 completed the formalization of Fermat's Last Theorem in two days, and Anthropic released two flagship models at once

Address: https://hexagonmath.org/2611.00174v1

Protein Folding

Adeno-associated Virus (AAV) is a very small and harmless virus to humans. Its protein shell, known as the Capsid, can be used to deliver gene therapies. The protein structure of the capsid is very important. For example, it affects whether the gene therapy will be recognized and blocked by the patient's immune system. However, the structure of the capsid is also quite complex.

Existing protein folding models like AlphaFold can predict the structure of individual proteins, but a viral capsid is composed of dozens of proteins, making the overall structure much more complex.

Claude Mythos 5.2 found a new method based on two open-source protein folding models that can accurately predict the structure of the entire capsid. Specifically, it wrote code to let these two models handle protein chain groups with overlapping regions separately, and then integrated the predicted results of each part using the known symmetry information of the capsid to ultimately reconstruct the complete capsid structure.

This method has achieved good results on multiple natural and artificially designed capsids. In our tests, out of the 34 capsids that Claude attempted to model, 28 received accurate structural predictions.

Mathematics

Mathematical formalization refers to converting mathematical proofs into a language that computers can check step by step. This work can help mathematicians verify research results and enhance confidence in the correctness of proofs without relying entirely on manual step-by-step checks. This is especially important when AI models provide results that are difficult for humans to verify directly.

However, mathematical formalization is often very time-consuming. If AI can assist in this work, mathematicians can devote more time to exploring new mathematical ideas.

Previously, Anthropic used an unreleased research version of Claude to formalize Fermat's Last Theorem, which took 11 days. This work relied on a specially designed execution framework and required human researchers to provide guidance.

In contrast, a team of 1000 Fable 5.2 agents completed the formalization of the entire theorem in just 30 hours, with almost no human intervention.

Claude Fable 5.2 completed the formalization of Fermat's Last Theorem in two days, and Anthropic released two flagship models at once

Fable 5.2 started with very limited prompts and independently established a collaboration mechanism among the agents. Throughout the process, it did not use any mathematical reference materials, relying only on a list of intermediate theorem statements to assess task progress.

Compared to our first formalization process, this task consumed only one-third of the action tokens, and the length of the generated proof was reduced to one-seventh of the original.

The following figure shows the performance of different models: under the same configuration using 1000 agents, Opus 5.5 took 18 hours to complete only 26% of the formalization work. Even if it continued to run for 30 hours, progress was minimal.

Claude Fable 5.2 completed the formalization of Fermat's Last Theorem in two days, and Anthropic released two flagship models at once

Hardware and Robotics

Many potential values of AI, such as accelerating laboratory research, depend on its ability to take action in the physical world. Claude Fable 5.2 continues the trend of increasing practicality of the Claude series models in the real world. It has shown good capabilities in tasks such as hardware device design and has made significant progress in controlling physical robots.

Anthropic specifically tested Fable 5.2's hardware engineering capabilities by giving it a brief prompt to design and manufacture three functional electronic devices. Each device needed to have an independent circuit board, a 3D printed shell, and accompanying software. The final three devices were: a home voice assistant hub, a USB speakerphone, and a handheld virtual pet device.

Claude independently wrote the product specifications, selected electronic components available from U.S. distributors, designed and completed four-layer circuit board layouts for each device, created shell models using computer-aided design (CAD), and wrote the software required for the devices. With almost no human intervention, each circuit board design was completed in less than a day, reaching a state ready for manufacturing.

After the circuit board manufacturing was completed, Claude further guided the company's engineers in device testing and assembly. Ultimately, all three devices worked normally without any modifications to the circuits.

Robotics

Compared to previous Claude models, Fable 5.2 has shown a significant improvement in controlling physical robots. Even when faced with tasks for which it had never received specialized training, it demonstrated strong operational capabilities.

The video below shows a robot controlled by Fable 5.2 completing multiple tasks, including solving a Rubik's Cube, unscrewing bottle caps, and drawing a robot picture.

In the public RoboDojo benchmark tests, researchers used camera images to have AI control a simulated dual-arm robot to complete tasks. Fable 5.2 not only outperformed other leading models using the same testing framework but also surpassed most AI systems specifically developed for robot control.

Claude Fable 5.2 completed the formalization of Fermat's Last Theorem in two days, and Anthropic released two flagship models at once

RoboDojo Evaluation Results

Research and Analysis

Early testers found that when using Claude Fable 5.2 for research and analysis, it could handle more ambiguous and larger data tasks compared to previous models, demonstrating better judgment when faced with complex and unclear problems, especially excelling in tasks requiring abstract reasoning and solution design.

The proprietary trading firm Jump Trading tested Fable 5.2 using some agents simulating their quantitative research tasks. During the testing process, they connected the model to a system similar to their internal research framework.

In one of the tests, Fable 5.2 almost independently took on all necessary roles, including code writer, task coordinator, planner, and task submitter. Previously, when using other models, the team often needed to invest a lot of time manually adjusting the division of labor among different roles to fit the strengths and weaknesses of each model.

Safety and Alignment

No single test can fully predict how a high-capacity model will perform after deployment. Therefore, before official release, we will comprehensively use multiple testing methods and collect different types of evidence.

For the models underlying Claude Fable 5.2 and Claude Mythos 5.2, the safety work conducted by Anthropic before deployment includes: alignment training and testing, designing new evaluations to test the model's tendency to take harmful actions, establishing stronger anti-abuse mechanisms, and expanding the scope of behavior monitoring.

Before the official release, the model also underwent testing by external evaluation agencies, including the UK AI Safety Research Institute, Frontier Design, and METR.

Alignment Testing

Anthropic conducted a series of evaluations to examine the alignment of Claude Fable 5.2 and Claude Mythos 5.2 from a behavioral perspective, focusing on the extent to which they exhibit unexpected behaviors and behaviors that may pose safety risks.

Based on multiple evaluation results, both models have a low risk of exhibiting unsafe behavior in high-risk scenarios, generally on par with other recently released models.

Anthropic tested Claude Mythos 5.2 using automated behavior audits. It should be noted that Mythos 5.2 uses the same underlying model as Fable 5.2 but lacks some of the latter's cybersecurity and biosafety protections. This automated audit covered approximately 1900 different scenarios to test the model's behavior in various situations.

On most evaluation metrics, Claude Mythos 5.2 performed at par or better than Opus 5.5, Sonnet 5.5, and Claude Mythos 5.1. For example, in most evaluations involving misleading users, it outperformed these three models. Specifically, whether it was fabricating input information, falsely claiming to have completed tasks, omitting important information, or inadequately describing poor work quality or being perfunctory, the incidence of these issues was lower for Claude Mythos 5.2.

Claude Fable 5.2 completed the formalization of Fermat's Last Theorem in two days, and Anthropic released two flagship models at once

During the testing process, Anthropic paid special attention to behavioral patterns involved in recent cybersecurity and alignment incidents that have drawn attention. For example, compared to Opus 5.5, Sonnet 5.5, and Mythos 5.1, Claude Mythos 5.2 exhibited fewer instances of biased or motivation-driven reasoning. Such reasoning could lead the model to dismiss evidence indicating that it is not in a simulated environment, thereby taking actions without a full understanding of the real-world consequences.

Anthropic also tested the model's tendency to attempt to escape the sandbox through various assessments. In the two assessments previously used, Claude Mythos 5.2 attempted sandbox escape significantly fewer times than all other models we tested.

Additionally, Anthropic designed a dedicated adversarial test covering over 1,800 scenarios. Most scenarios could not be completed without escaping the sandbox, and the testing environment itself was deliberately set with some security weaknesses. In this test, Claude Mythos 5.2 used credentials or files placed outside the sandbox in 3.1% of the scenarios.

This percentage is roughly equivalent to the 3.3% of Opus 5.5, lower than the 5.1% of Sonnet 5.5, and far below the previous two versions of Claude Mythos: Mythos 5 had a rate of 39%, and Mythos 5.1 had a rate of 21%.

Anthropic also found that by using other methods, such as explaining the specific environment of the sandbox to the model and providing tools for reporting security weaknesses, the attempts of Claude Mythos 5.2 to escape could be further reduced.

Meanwhile, Anthropic is continuously improving assessment methods to make the testing environment closer to real-world scenarios and further clarify what issues existing assessments may have overlooked.

For more details about the model, please refer to the 264-page model card:

Claude Fable 5.2 completed the formalization of Fermat's Last Theorem in two days, and Anthropic released two flagship models at once