Anthropic has introduced two new AI models, Claude Fable 5 and Claude Mythos 5, positioning the pair as its most capable systems yet while limiting access to the more powerful variant.
The company said Fable 5 is a Mythos-class model that has been made safe for general use. It described the model as outperforming every previous Claude release on nearly all of its tested benchmarks, with especially strong results in software engineering, knowledge work, vision, scientific research and other long-running tasks.
At the same time, Anthropic emphasized that the release comes with safety controls. For some topics, the system will not answer directly and will instead route prompts to Claude Opus 4.8, the company’s next-most-capable model. Anthropic said those safeguards were intentionally set conservatively so the model could be released quickly and with caution, even though that means some harmless requests may be blocked. The company said the protections trigger in fewer than 5% of sessions on average.
Alongside Fable 5, Anthropic is offering Claude Mythos 5 to a smaller group of cyber defenders and infrastructure providers. Mythos 5 uses the same underlying model as Fable 5, but with fewer restrictions in certain areas. The first deployment will come through Project Glasswing, a government collaboration that Anthropic says is an upgrade to Claude Mythos Preview.
The company said Mythos 5 has the strongest cybersecurity capabilities of any model it has seen so far. It plans to widen access later through a broader trusted-access program.
Anthropic also framed the launch as a continuation of its effort to make advanced AI more broadly available while maintaining controls around potentially dangerous uses. The company said the models could be useful in areas ranging from cyber defense to life sciences, where it claims they are already helping researchers generate new hypotheses and move faster on therapeutic development.
In its release materials, Anthropic highlighted several examples of early testing. It said Stripe reported that Fable 5 compressed months of engineering work into days and handled a large-scale migration in a Ruby codebase that would otherwise have taken a team more than two months.
The company also pointed to gains in analytical work, saying the model performed strongly on finance and trading evaluations that tested document reasoning, chart interpretation and root-cause analysis. In vision tasks, Anthropic said Fable 5 can extract numbers from scientific figures and reconstruct working web apps from screenshots. It also said the model can complete games such as Pokémon FireRed using only visual input.
Another area of emphasis was memory and long-context handling. Anthropic said Fable 5 stays on task across millions of tokens and can improve its own output by referring back to notes it has created during a project.
Mythos 5 was presented as more than a coding tool. Anthropic said its internal protein-design experts used it to speed parts of drug discovery by about tenfold. The company also said researchers preferred Mythos 5’s molecular biology hypotheses in blinded comparisons about 80% of the time and that one of those ideas was later supported by an independent lab. In genomics, Anthropic said the model spent more than a week on largely autonomous research and produced a smaller machine-learning system that outperformed a recent model published in Science.
Anthropic said both models are priced at $10 per million input tokens and $50 per million output tokens, which it described as less than half the price of Claude Mythos Preview. The company said the launch is part of a broader plan to make advanced AI available quickly and safely.
The announcement also included early alignment results. Anthropic said Mythos 5 showed a low level of misaligned behavior, similar to Opus 4.8, and said Fable 5 should have a comparable profile because the two systems share the same base model.
The launch comes as AI labs continue to push models toward longer autonomy, higher coding performance and more specialized scientific use. Anthropic’s decision to pair a public release with a restricted-access version reflects the tension between wider availability and the risks of deploying highly capable systems.