Anthropic has publicly released Claude Fable 5, its newest Mythos-class AI model, making it the most capable system the company has offered for broad use to date.
The launch provides developers and subscription customers with access to Fable 5 via the Claude API and across several Anthropic subscription plans.
In parallel, Anthropic is offering a variant called Claude Mythos 5. Access to this model is restricted and limited to selected cyber defenders, infrastructure providers, and biology researchers through a trusted access programme.
Anthropic reports that Fable 5 outperforms all previous Claude models on nearly every tested benchmark. The company highlights improvements in software engineering automation, knowledge work, vision tasks, and scientific research.
The Claude developer claims that the model’s advantage grows for longer and more complex tasks. Both Fable 5 and Mythos 5 can operate autonomously for longer than prior Claude models.
To reduce risks of misuse, Anthropic has added safeguards to Fable 5’s deployment. Some queries, especially those related to sensitive topics such as cybersecurity, are automatically redirected to the Claude Opus 4.8 model.
According to Anthropic, the safeguards are conservative and trigger in less than 5% of user sessions. This means some harmless requests may also be blocked. The company plans to update the system to reduce false positives and improve the balance between safety and accessibility as future models are developed.
Claude Mythos 5 shares the same model architecture as Fable 5 but has certain safeguards lifted. It is initially available to a small group through Project Glasswing in partnership with the US government.
The limited access is designed to address high-security cybersecurity use cases and support controlled research. Anthropic intends to expand access to Mythos 5 gradually through a trusted access process for selected organisations in cybersecurity and life sciences.
Anthropic stated: “The capabilities of models like Fable 5 and Mythos 5 have the potential to do profound good for the world. We’ve seen the beginnings of this in Project Glasswing, where the models have helped cyber defenders secure critically important software.
“We’ve also seen it in life sciences research, where the models are positing novel hypotheses and speeding up the development of new therapeutics.”
Pricing for both Fable 5 and Mythos 5 is set at $10 per million input tokens and $50 per million output tokens, which is less than half the cost of the earlier Claude Mythos Preview.
Fable 5 is accessible for developers via the API and for customers on Pro, Max, Team, and seat-based Enterprise plans until 22 June at no extra charge. After this, usage requires credits.
Anthropic may extend the no-charge window if capacity allows and aims to restore Fable 5 as a standard feature on subscription plans when possible. The company says it will update users about any access or capacity changes.
In technical testing, Stripe reported using Fable 5 to migrate a 50-million-line Ruby codebase in a single day, compared to over two months with a team.
Cognition’s FrontierCode evaluation gave Fable 5 the highest score for maintainable agentic coding. The model ranked highly on the Hebbia Finance Benchmark and IMC’s trading-analysis tests.
For vision tasks, Fable 5 extracted scientific data and rebuilt web applications from screenshots. It maintained performance in long tasks involving millions of tokens. In game simulations, Fable 5 achieved better results than previous models when using persistent memory.
Internal studies with Mythos 5 showed faster protein design and the generation of new biology hypotheses for experimental evaluation.
In genomics, Mythos 5 analysed single-cell data from 138 animal species and produced a machine learning model that outperformed a recent Science publication. Anthropic reports that misaligned behaviour in both models was low and similar to Claude Opus 4.8.
Recently, Anthropic closed a Series H round, raising $65bn and pushing its post-money valuation to $965bn. The company also filed a draft registration statement on Form S-1 to the US Securities and Exchange Commission (SEC) on a confidential basis for a potential initial public offering (IPO).
