Claude Fable 5 Brings Mythos AI to Public Users

Anthropic has released Claude Fable 5, the first publicly available version of its Mythos-class artificial intelligence (AI) system. The launch brings Anthropic's most advanced long-running autonomous capabilities to developers and businesses. The launch provides access to technology that had previously been restricted under Project Glasswing, Anthropic's program for cybersecurity and critical infrastructure partners.

The company calls Fable 5 its most capable publicly available model. According to the announcement, it is "state-of-the-art on nearly all tested benchmarks of AI capability," spanning software engineering, vision, knowledge work, and scientific research. Anthropic also says Fable 5 performs better as tasks become longer and more complex, working independently for hours while maintaining context.

Claude Fable 5 is Designed for Long-Horizon Work

Fable 5 shares the same underlying model as Claude Mythos 5, but includes safeguards that restrict certain cybersecurity, chemistry, biology, and model-distillation requests. When those requests are detected, they are automatically routed to Claude Opus 4.8.

Anthropic reported that the safeguards trigger in fewer than 5% of sessions on average.

Benchmark results released by the company place Claude Fable 5 among the top-performing AI systems. It scored 80.3% on SWE-Bench Pro, 29.3% on FrontierCode Diamond, 72.9% on CursorBench, 57.9% on Databricks OfficeQA Pro, and 85% on OS-World Verified. Anthropic further pointed out that the model remains effective on tasks that run for hours or days, unlike systems that plateau when given additional compute.

Early testing has produced examples that extend beyond traditional coding assistance. University of Pennsylvania professor and AI researcher, Ethan Mollick, noted that Fable 5 "outperformed basically every other public model" he had used, while demonstrating the ability to work on complex, multi-page specifications for up to 12 hours.

Using a single prompt in Claude Code, Mollick generated several playable games, ranging from the exploration title Strata and a Snake-style arcade game, to Duino, a project inspired by Rainer Maria Rilke's Duino Elegies. He also used the model to create a detailed isochronic travel-time map.

Safety Measures and Enterprise Rollout

Anthropic paired the release with a new data policy requiring 30-day retention of inputs and outputs for Mythos-class models. The company clarified that the data will be used for safety monitoring and classifier improvements, not for model training.

Alongside Fable 5, Anthropic launched Claude Mythos 5 for a limited group of cybersecurity organizations and researchers. The company noted Claude Mythos Preview has already helped identify more than 10K high- and critical-severity software vulnerabilities and is being expanded through a trusted-access program.

Anthropic cited several large-scale deployments during testing. Stripe reportedly used Fable 5 to complete a migration across a 50M-line Ruby codebase in a single day, a project the company estimated would otherwise require a team of engineers working for more than two months. Internal research teams also used Mythos 5 to accelerate protein-design workflows and generate novel molecular biology hypotheses that have advanced to experimental evaluation.

Fable 5 is available through the Anthropic API, Amazon Bedrock, Claude Code, Google Cloud's Vertex AI, Microsoft Foundry, GitHub Copilot, Databricks, Cursor, Devin, Lovable, v0, Vercel AI Gateway, and Augment Code. Pricing is set at $10 USD per million input tokens and $50 per million output tokens, with subscription users receiving temporary access before usage credits become mandatory after June 22, 2026.

The release marks Anthropic's push beyond chatbots toward AI systems that can independently execute complex software, research, and analytical tasks.