Anthropic launches Fable 5.2 and Mythos 5.2 with scientific and coding upgrades

Anthropic PBC today unveiled Claude Fable 5.2 and Claude Mythos 5.2, the two most powerful large language models to date.

The LLMs are identical in terms of their core competencies. The difference is that Fable 5.2 blocks certain use cases in biology and cybersecurity, while Myth 5.2 has looser guardrails. The latter model is only available to trusted researchers through two programs with limited access.

Anthropic tested Fable 5.2 against its predecessor in 14 artificial intelligence benchmarks. The model outperformed Fable 5.1’s score by more than 10% in most tests. It’s particularly good for scientific research: Fable 5.2 achieved 74.6% on Terminal Bench Science 0.1, a 22% improvement over Fable 5.1.

Earlier this year, Anthropic used an unreleased LLM to generate a computer-verifiable version of a famous mathematical proof. The project took 11 days. According to Anthropic, Fable 5.2 reduced workflow to 30 hours by deploying 1,000 agents to parallelize calculations. In a separate project, it answered two open questions about a theoretical physics concept related to black holes.

The company also tested Mythos 5.2, the invite-only version of Fable 5.2. The model was asked to examine the protective shells of harmless viruses, called AAVs. AAV envelopes, or capsids as they are known in scientific parlance, are used to deliver drugs in some therapies.

Myth 5.2 accurately modeled the structure of 28 of the 34 AAV capsids examined. The result is remarkable because capsids are made up of dozens of different proteins. Other cutting-edge models optimized for biology research, notably Google DeepMind’s AlphaFold2, can only model the structure of individual proteins. The creators of the latter model won half of the 2024 Nobel Prize in Chemistry.

Fable 5.2 includes guardrails that block most research biology questions and many cybersecurity-related prompts. When users enter a request that falls into one of these categories, the model sends it to Opus 4.8 or Opus 5. According to Anthropic, Fable 5.2 is 55% less likely to incorrectly block safe prompts than Fable 5.1.

The model outperforms its predecessor by more than 13% in two coding benchmarks called FrontierCode 2.0 FrontierSWE 2.0. According to Anthropic, one of its engineers put the model’s programming skills to the test by tasking it with developing an operating system. Fable 5.2 spawned 1,000 agents who completed the task in about eight days. The operating system they created can run on a laptop, provides network access, and has a Claude sidebar.

Anthropic sees customers using Fable 5.2 and Mythos 5.2 to accelerate highly complex, open technical tasks. The company says that with Opus 5.5, an LLM launched in July, more everyday workflows can be automated more efficiently. Users can also combine the models by having Opus 5.5 ask Fable 5.2 or Mythos 5.2 for advice when a bottleneck occurs.

The price for Fable 5.2 and Mythos 5.2 is $8 per million input tokens and $40 per million output tokens. This makes them 20% cheaper than Anthropic’s previous models. Both LLMs offer a high-performance mode that generates quick responses 2.5 times faster than the standard level at twice the price.

Image: Anthropocene