New model released: Claude Fable 5 and Claude Mythos 5

On June 9, 2026, Anthropic introduced two new models: Claude Fable 5 and Claude Mythos 5. Both belong to a new model class called "Mythos", positioned above the existing Opus line, and according to Anthropic they set a new state of the art on nearly all tested benchmarks.

The key facts

• Two models, one core: Fable 5 and Mythos 5 are based on the same underlying model. Fable 5 is intended for general public use and runs with additional safety classifiers for areas such as cybersecurity and biology. Mythos 5 is initially aimed only at a small circle of trusted partners (such as cyber-defense teams via Project Glasswing) and has these safeguards lifted in part.
• Strong results in practice: Payments company Stripe reported that the model completed a codebase migration in a single day that would otherwise have taken a team more than two months. On Cognition's FrontierCode evaluation for demanding coding tasks, Fable 5 achieved the highest score of any tested model.
• Pricing: Both models cost $10 per million input tokens and $50 per million output tokens, which Anthropic says is less than half the price of the previous Mythos Preview.
• Availability: Claude Fable 5 is available via the Claude API, Amazon Bedrock, Google Cloud, Microsoft Foundry, and within Claude, Claude Code, and Claude Cowork. Claude Mythos 5 remains limited to select partners for now.

Benchmark comparison

At launch, Anthropic published a comparison table positioning Fable 5 and Mythos 5 against other leading models. According to Anthropic, Fable 5's biggest lead shows up precisely on long, complex tasks: the more demanding the task, the larger the gap to previous Claude models.

Anthropic also reports top scores in other areas: on Hebbia's Finance Benchmark for demanding analytical tasks, Fable 5 achieved the highest score of any model, with substantial gains in document-based reasoning as well as chart and table interpretation. Trading firm IMC reported that the model passed its trading-analysis evaluations almost across the board, from factual lookup to conceptual reasoning to root-cause analysis. On vision tasks, Anthropic considers Fable 5 a new state of the art: earlier Claude models struggled with the game Pokémon FireRed even with additional helper tools, while Fable 5 completed it using only a minimal, vision-only setup.

FrontierCode and agentic coding

For software engineering, Anthropic specifically points to Cognition's FrontierCode evaluation, which tests whether models can solve demanding coding tasks to the standard of high-quality production codebases. According to Anthropic, Fable 5 achieves the highest score of any tested frontier model there, even at medium effort.

Several early testing partners highlight the same point: Cognition CEO Scott Wu called Fable 5 the top-scoring model on FrontierBench, with particular strength in long-horizon reasoning and the ability to generalize to unfamiliar tools without prior setup. Cursor CEO Michael Truell described it as opening up a new class of long-horizon problems that were simply out of reach for earlier models.

A brief note on availability since launch

Three days after launch, the US government briefly imposed export controls that temporarily suspended access to both models worldwide. Since July 1, 2026, Fable 5 and Mythos 5 have been regularly available again.