Anthropic launched Claude Fable 5.2 and Claude Mythos 5.2 for complex coding, research, and cybersecurity tasks. These models generate outputs 30% faster than previous versions. While the general public can use Fable 5.2, Mythos 5.2 is restricted to approved organizations. Anthropic also lowered prices to make these advanced tools more accessible.

Anthropic has unveiled two new AI models -- Claude Fable 5.2 and Claude Mythos 5.2 -- optimized for complex coding, scientific research, engineering and cybersecurity tasks. The models announced yesterday offer developers, researchers and organisations tools to solve more complex problems that demand advanced reasoning and autonomous operation. The new models can be accessed via Anthropic's platforms, with Mythos available only to approved organisations in the cybersecurity and life sciences sectors.

According to Anthropic, the models can generate outputs up to 30% faster than their predecessors, Claude Fable 5.1 and Claude Mythos 5.1. The company has also lowered their prices to $8 per million input tokens and $40 per million output tokens. For setups with more tokens, the optional Fast Mode can boost the number of tokens generated by up to 2.5 times, but at a higher cost.

The two models share the same underlying architecture but have different safety restrictions, Anthropic says. Cybersecurity and biology protections are incorporated into Claude Fable 5.2, which the general public can use. Claude Mythos 5.2 is designed for trusted organisations engaged in cutting-edge cybersecurity and biomedical research.

According to the company, Fable 5.2 has surpassed Claude Opus 5.5 on most of the benchmarks it has been tested on. On FrontierCode 2.0, it achieved 49.1% at the X-high effort level, while Opus 5.5 achieved 45.5%. It also achieved a high success rate on the RoboDojo robotics benchmark, at 64.9%.

AI model demonstrates advanced coding and scientific research

During the pre-release testing, Anthropic emphasized several applications. The company said Fable 5.2 helped complete a critical production service migration from Python to Rust, making the service more efficient and approximately 26 times faster to start.

In another test, the model has assisted in developing a working operating system from scratch in eight days, with the help of over 1,000 AI agents.

Anthropic also stated that a team of 1,000 Fable 5.2 agents formalized the proof of Fermat's Last Theorem in 30 hours, with very little human intervention. The company said the system also solved two previously open problems related to black hole physics.

Anthropic strengthens AI safety and cybersecurity protections

The new versions continue to emphasize safety. Anthropic states that it has undergone extensive pre-release testing, including assessments of potentially dangerous behaviours and tests of how the bot could exit controlled computing environments.

The company said that Fable 5.2 had about 55% fewer unneeded safety blocks than the previous version. Mythos 5.2 also showed improvements in the internal test to identify software vulnerabilities.

Availability and pricing

Claude Fable 5.2 can be used on all Anthropic platforms, including Amazon Web Services, Google Cloud and Microsoft Azure. It's also available to developers via the Claude Platform.

Claude Mythos 5.2 is currently limited to eligible organisations under Anthropic's Cyber Verification Program and Life Sciences Verification Program. It also has a Fast Mode, which is restricted to qualified organisations.