Anthropic launches Claude Fable 5, unveils restricted-access Claude Mythos 5

Anthropic said Claude Fable 5 can work autonomously on longer and more complex tasks than previous Claude models.Anthropic on Wednesday launched Claude Fable 5, a new Mythos-class AI model that the company said is its most capable model made available for general use, while also unveiling Claude Mythos 5 for a limited group of cybersecurity and infrastructure partners.According to the company’s statement, Claude Fable 5 delivers state-of-the-art performance across software engineering, knowledge work, vision, scientific research and other benchmarks.

To address safety concerns, Anthropic has introduced safeguards that route certain cybersecurity, biology, chemistry and distillation-related queries to its next-most-capable model, Claude Opus 4.8.The safeguards have been tuned conservatively and trigger in less than 5 per cent of sessions on average.Claude Mythos 5, which uses the same underlying model as Fable 5 but with some safeguards removed, is initially being deployed through Project Glasswing in collaboration with the US government.

Anthropic said Mythos 5 has the strongest cybersecurity capabilities of any model in the world and will later be expanded through a broader trusted access programme.The company has priced both models at $10 per million input tokens and $50 per million output tokens.Focus on software engineering, research and long-context tasksAnthropic said Claude Fable 5 can work autonomously on longer and more complex tasks than previous Claude models.In software engineering, the company cited testing by Stripe, which reported that Fable 5 completed a codebase-wide migration in a 50-million-line Ruby codebase in a day.

On Cognition’s FrontierCode evaluation, Anthropic said Fable 5 achieved the highest score among frontier models.For knowledge work, the model recorded the highest score on Hebbia’s Finance Benchmark for senior-level reasoning, according to the company.Anthropic also highlighted improvements in vision capabilities, stating that Fable 5 can extract information from scientific figures and rebuild web application source code from screenshots.The model also demonstrated stronger long-term context and memory capabilities, including improved performance in the game Slay the Spire when given access to persistent file-based memory.Mythos 5 used for drug design and genomics researchAnthropic said internal protein design experts used Mythos 5 to accelerate parts of the drug discovery process by around ten times.According to the company, Mythos 5 was able to perform tasks such as selecting binding sites, running protein design tools, and recovering from failures without human assistance.

Nine of 14 protein targets studied yielded strong drug-design candidates that are currently under investigation.The company also said Mythos 5 generated novel molecular biology hypotheses that Anthropic scientists preferred around 80 per cent of the time compared with hypotheses produced by Opus-class models.In another research project, Anthropic said Mythos 5 conducted genomics research over more than a week of largely autonomous work, assembling single-cell data from millions of cells across 138 animal species and training a custom machine learning model.New safeguards and data retention policyAnthropic said Mythos-class models present significant risks due to their advanced cybersecurity and biology capabilities.To mitigate misuse, Fable 5 uses AI-based safety classifiers that detect potentially harmful requests.

When such requests are identified, responses are automatically handled by Claude Opus 4.8 instead.The company said cybersecurity safeguards are designed to cover exploit development, offensive cyber tasks and agentic hacking activities.

For biology and chemistry, Fable 5 currently falls back to Opus 4.8 for most related requests.Anthropic also announced a new data retention policy for Mythos-class models, requiring 30-day retention of business customer traffic for safety monitoring purposes.

The company said the data will not be used to train future Claude models and will be deleted after 30 days in almost all cases.AvailabilityClaude Fable 5 is available globally from today.

It remains restricted to Project Glasswing partners and will later be made available to select biology researchers through a trusted access programme.The company has priced both models at $10 per million input tokens and $50 per million output tokens.

Developers can access Claude Fable 5 through the Claude API.Anthropic said it expects demand for Fable 5 to be strong and potentially difficult to forecast.

The model is available immediately on the Claude API and consumption-based Enterprise plans.For subscription users, Fable 5 will be included at no additional cost on Pro, Max, Team and seat-based Enterprise plans through June 22.

From June 23, continued use of the model will require usage credits, although the company said it may extend the complimentary access period if capacity permits.Anthropic added that it intends to eventually make Fable 5 a standard feature across subscription plans once sufficient computing capacity becomes available and said users would be informed in advance of any changes to availability.