Anthropic Launches Claude Fable 5.2 and Mythos 5.2

0

Anthropic has introduced two advanced Claude AI models aimed at tackling complex technical challenges, with improvements in coding, scientific research, cybersecurity and reasoning.

Anthropic Claude Fable 5.2 and Mythos 5.2 AI models launch announcement.
Anthropic introduces two advanced Claude AI models with faster performance and expanded technical capabilities. (AI Generated Image)

Artificial intelligence company Anthropic announced Claude Fable 5.2 and Claude Mythos 5.2 on November 9, expanding its AI lineup with models designed for demanding technical work. The new releases share the same underlying model but differ in their security protections and availability.

The launch gives developers and researchers access to upgraded AI capabilities while maintaining restrictions on certain high-risk applications. Anthropic says the models can handle more complicated tasks and generate responses faster than their previous versions.

Two Claude Models With Different Access Levels

Claude Fable 5.2 is the generally available version, designed for software engineering, advanced reasoning, robotics and scientific applications. Users can access it through Claude applications and supported developer platforms.

Claude Mythos 5.2 uses the same underlying model but has different cybersecurity and biology safeguards. It is intended for approved organizations working on cybersecurity and biomedical research rather than general public use.

Organizations can apply for Mythos access through Anthropic's Cyber Verification Program and Life Sciences Verification Program. Enterprise customers can also access Mythos-powered security capabilities through Claude Security.

This separation allows Anthropic to offer advanced technical capabilities while placing additional controls on applications that could create cybersecurity or biological risks.

Faster Responses and Lower API Prices

Performance and pricing are two important parts of the announcement. Anthropic says Fable 5.2 and Mythos 5.2 can generate outputs up to 30% faster than their respective 5.1 versions.

The company has also reduced standard API pricing. Both models cost $8 per million input tokens and $40 per million output tokens, compared with $10 and $50 per million tokens for the previous generation.

Developers who need faster processing can choose an optional Fast Mode. This option costs $16 per million input tokens and $80 per million output tokens, with token generation speeds of up to 2.5 times the standard rate.

These pricing changes make the new models less expensive at standard rates, although actual costs depend on how much input and output developers use.

Anthropic Reports Improvements in AI Benchmarks

Anthropic published several benchmark results to demonstrate the models' technical capabilities. In FrontierCode 2.0, Claude Fable 5.2 achieved a score of 49.1% at Xhigh effort. It also scored 69.9% on FrontierSWE v2 and 51.4% on AutomationBench.

The model recorded 74.2% on Humanity's Last Exam and 64.9% on the RoboDojo Sim evaluation. These tests cover different areas, including coding, business workflows, multidisciplinary reasoning and simulated robotics.

For cybersecurity, Anthropic reported that Mythos 5.2 found more vulnerabilities in its internal CVE Rediscovery evaluation than its previous Mythos version. The company said the new model identified 14.4% more vulnerabilities than Mythos 5.1 in that evaluation.

These figures come from Anthropic's own published testing. Benchmark scores provide useful comparisons, but they do not guarantee that a model will deliver the same results in every real-world situation.

AI Agents Help Formalize Fermat's Last Theorem

Some of Anthropic's most notable claims involve mathematical research. During prerelease testing, the company used a team of 1,000 Fable 5.2 agents to formalize the proof of Fermat's Last Theorem in approximately 30 hours.

Mathematical formalization involves translating a proof into a computer-checkable format. The process allows mathematical statements to be verified systematically rather than relying entirely on manual examination.

Anthropic compared this experiment with an earlier attempt using an unreleased Claude model, which took 11 days. The company said the newer system completed the work with minimal human intervention.

Anthropic also reported progress on two previously unresolved theoretical physics problems involving black holes. These results came from the company's own research experiments and have not been presented as independently reproduced achievements.

New Developments in Electronics and Robotics

Fable 5.2 was also tested on practical engineering tasks. Anthropic reported that the model helped design three electronic devices from scratch: a home voice-assistant hub, a USB speakerphone and a handheld virtual pet.

The company worked with electronics company Diode on manufacturing and testing the devices. The experiments were intended to examine how AI could contribute to hardware development beyond generating software and answering technical questions.

Anthropic also reported improvements in simulated robotics tasks, suggesting that the model can contribute to work involving physical systems. However, performance in simulations does not automatically establish equivalent performance in real-world environments.

Safety Measures Remain a Major Consideration

The release includes additional safety testing as Anthropic expands the capabilities of its AI systems. The company says it conducted extensive prerelease evaluations covering alignment and potentially concerning behavior, including attempts to escape controlled computing environments known as sandbox escapes.

Anthropic identified external evaluators involved in its testing, including the UK AI Security Institute, Frontier Design and METR.

Fable 5.2 includes safeguards intended to block high-risk cybersecurity activities and certain biology-related requests. Mythos 5.2, meanwhile, is restricted to organizations that meet Anthropic's verification requirements.

Anthropic also reported that Fable 5.2 produces fewer unnecessary safety blocks than Fable 5.1. However, the company acknowledges that safety evaluations cannot anticipate every possible failure or misuse scenario.

Where Developers Can Access Fable 5.2

Claude Fable 5.2 became generally available on November 9, 2026. Users can access it through Claude applications, while developers can use the model through the Claude Platform and supported cloud services.

The model is also available through Amazon Web Services, Google Cloud and Microsoft Azure. Developers using the Claude API can identify the model with the name claude-fable-5-2.

Mythos 5.2 follows a more restricted distribution approach. Cybersecurity and life sciences organizations must qualify through the relevant verification programs to obtain access to its capabilities.

What the Launch Means for Anthropic

The release comes as competition among major AI developers continues to expand beyond general-purpose chatbots toward systems capable of completing complex technical work.

Anthropic, led by CEO Dario Amodei, has increasingly focused on AI applications involving software development, scientific research and enterprise security. The company's latest models reflect that direction by combining advanced reasoning capabilities with different levels of access and safeguards.

Anthropic's proposed initial public offering is another important development in the company's business history. It filed a registration statement with the U.S. Securities and Exchange Commission on October 26, 2026. However, the model announcement and the company's IPO process are separate developments.

For developers, researchers and businesses, the immediate significance of Fable 5.2 and Mythos 5.2 will depend on how effectively the models perform outside company testing. Their real-world usefulness, reliability and safety will become clearer as more organizations work with them.

Post a Comment

0 Comments
* Please Don't Spam Here. All the Comments are Reviewed by Admin.
Post a Comment (0)
To Top