Anthropic launches Claude Fable 5 and Claude Mythos 5
- Maxime Hiez
- Anthropic
- 10 Jun, 2026
Introduction
Anthropic announces the launch of Claude Fable 5 and Claude Mythos 5, two models built on the same foundation, with capabilities surpassing those of any previously released model. Fable 5 is generally available, while Mythos 5 remains restricted to a limited circle of authorized partners.
Two models, one shared foundation
Claude Fable 5 and Claude Mythos 5 share the same underlying model, referred to as Mythos-class by Anthropic.
- Claude Fable 5 : Generally available variant with active safety mechanisms. Available via the API and Claude subscriptions.
- Claude Mythos 5 : Same underlying model with certain restrictions lifted in specific domains. Initially deployed via Project Glasswing for cybersecurity professionals and infrastructure providers. Access remains limited to trusted partners authorized by Anthropic.
This distinction marks an evolution in Anthropic’s approach : offering a state-of-the-art model to the general public while providing a partially unrestricted version to qualified professionals.
Capabilities
Fable 5 achieves state-of-the-art performance on nearly all tested benchmarks. Key areas highlighted by Anthropic include :
- Software engineering : Reduction of months of work to days, according to partner feedback.
- Knowledge work : Financial analysis, complex reasoning across long contexts spanning millions of tokens with persistent memory.
- Vision : Data extraction from scientific figures, source code reconstruction from screenshots ; completion of Pokémon FireRed in vision-only mode, where previous models required complex auxiliary tools.
- Scientific research : Tenfold acceleration in drug design and protein research ; 9 out of 14 protein targets yielding strong drug candidates ; first Claude model preferred in 80% of cases in blinded comparisons against Opus-class models for molecular biology hypothesis generation.
Benchmarks
Fable 5 sets new records across several reference evaluations :
- FrontierCode (Cognition) : Highest score among frontier models, including at medium effort levels.
- FrontierBench : Highest score for frontier coding evaluation.
- Hebbia Finance Benchmark : Highest score for senior-level analytical reasoning.
- Analytics (Bloomberg) : First model to exceed 90% on their core analytics benchmark, a 10-point jump over Opus.
- Frontier physics : Results comparable to GPT-5.5 using one third of the reasoning tokens.
Customer testimonials
Several partners shared their experience with Fable 5 :
- Cursor : Describes Fable 5 as “state of the art” and reports access to categories of problems previously out of reach.
- GitHub : States that Fable 5 “exceeded previous benchmarks” on autonomous coding tasks.
- Stripe : Highlights the ability to complete full codebase migrations in a single day rather than months.
- Bloomberg : Notes strong performance on frontier physics research with reduced token consumption.
Safety mechanisms
Fable 5 includes classifiers that automatically redirect requests to Claude Opus 4.8 when they relate to :
- Offensive cybersecurity : Requests related to hacking or system exploitation.
- Dual-use biology and chemistry : Requests posing risks in sensitive biological research, including viral design, AAV engineering, or dangerous pathogen creation.
- Distillation : Attempts to extract model capabilities to train competing models.
These safeguards activate in fewer than 5% of sessions on average. Adversarial testing using 30 different jailbreak techniques conducted by external teams revealed no successful attacks.
Anthropic also applies a structured data retention policy for business customers : mandatory 30-day retention, no use for model training beyond safety purposes, logged human data access with strict deletion protocols.
Regarding Mythos 5, automated alignment testing reveals a level of misaligned behavior comparable to Claude Opus 4.8.
Pricing
Claude Fable 5 is available at :
- Input : 10$ / 1M tokens
- Output : 50$ / 1M tokens
This price represents less than half the cost of Claude Mythos Preview, the previous version available to partners.
Availability
- Claude Fable 5 : Available worldwide since June 9, 2026 via the Anthropic API and consumption-based plans. The rollout to Claude subscriptions occurs in phases through June 22, after which a usage credit model applies.
- Claude Mythos 5 : Access restricted to authorized partners under Project Glasswing. A dedicated trusted access program for biology research is also launching for selected researchers.
Conclusion
With Fable 5, Anthropic offers the general public a model from its Mythos-class lineup for the first time, a tier previously reserved for restricted circles. The distinction between the two variants illustrates a phased deployment strategy that exposes the model’s maximum capabilities while maintaining robust safeguards for general use.
Sources
Did you enjoy this post ? If you have any questions, comments or suggestions, please feel free to send me a message from the contact form.
Don’t forget to follow us and share this post.
