Anthropic launches Claude Fable 5.1 and Claude Mythos 5.1
- Maxime Hiez
- Anthropic
- 02 Sep, 2026
Introduction
Anthropic announces Claude Fable 5.1 and Claude Mythos 5.1, two variants of a single model succeeding Claude Fable 5 and Claude Mythos 5. Beyond the capabilities, the announcement mainly addresses the three most frequent customer complaints about the previous generation : price, data retention, and overly sensitive safeguards.
Check the Claude Fable 5 article HERE.
Two models, one shared foundation
Fable 5.1 and Mythos 5.1 are the same model, up to a level of safeguards.
- Claude Fable 5.1 : Generally available version, with safeguards active. Identifier claude-fable-5-1.
- Claude Mythos 5.1 : Same model, with relaxed safeguards for cybersecurity and life sciences. Access restricted to Anthropic’s trusted access programs.
Access to Mythos 5.1 now goes through two distinct programs :
- Cyber Verification Program : Intended for defensive security professionals, with reduced cyber safeguards.
- Life Sciences Verification Program : Built in partnership with the US government, it opens advanced biology capabilities to research professionals.
Both programs remain limited to a set of US organizations, with Anthropic stating that it is working to broaden access.
Benchmarks
Fable 5.1 is presented as the best model on the market for coding, knowledge work, and long-running problem solving. Anthropic insists on a qualitative point rather than the raw score : the model avoids shortcuts that create the appearance of a result, it says when it is blocked instead of declaring success, and it does not disable a failing test to make a build pass.
| Fable 5.1 | Fable 5 | Opus 5 | GPT-5.6 Sol | |
|---|---|---|---|---|
| Terminal-Bench-Science 0.1 | 52.6% | 24.7% | 29.0% | 22.4% |
| Terminal-Bench 4.0 | 55.8% | 42.0% | 52.3% | 37.3% |
| GDPval-AA v2 | 1853 | 1723 | 1824 | 1711 |
| OSWorld 2.0 (strict) | 41.7% | 36.1% | 39.6% | n/a |
| Humanity’s Last Exam | 60.9% | 57.8% | 56.6% | n/a |
| AutomationBench | 31.4% | 17.1% | 26.9% | 19.6% |
| CursorBench 3.2.0 | 73.4% | 70.5% | 70.0% | 67.2% |
Mythos 5.1, tested without production safeguards, reaches 60.9% on Terminal-Bench 4.0. Also worth noting, Fable 5.1 at low and medium effort levels achieves results similar to or better than Fable 5, at a much lower cost.
Customer testimonials
Early access partners report gains on cost as much as on quality :
- Jane Street Capital : States that Fable 5.1 “solves more of our coding problems than Fable 5 or Opus 5” and remains readable over long, multi-step tasks.
- Millennium : Reports that Fable 5.1 “was the first to find it” for a rare crash, about one in a million runs, that nobody had explained in four to five years.
- Cognition : Moves its Devin Opus 5 traffic to Fable 5.1 on launch day, the model having “matched or edged out Fable 5” in testing at a lower cost per task.
- Every : Sums up the model in one formula, “Fable-level intelligence, Opus-level price, Sonnet-speed”, at about twice the speed of Opus 5 using half as many tokens.
Pricing
The token pricing of Fable 5.1 is unchanged compared to Fable 5 :
- Input : 10$ / 1M tokens
- Output : 50$ / 1M tokens
- Cache reads : 0.25$ / 1M tokens
Cache reads are what changes everything, with a 75% reduction. On a typical workload, the overall cost drops by around 25% compared to Fable 5, and by up to around 45% on highly agentic workloads, where cache reads make up most of the bill.
Scientific research
Anthropic devotes a large part of the announcement to the scientific capabilities of both models, with experimentally validated results :
- Molecular design : Mythos 5.1 designed protein binders whose affinity is ten times higher than the best submissions to the Adaptyv Bio competitions on three targets. Its hit rate reaches nearly 50% across 12 targets, where 10 to 15% is the current norm.
- Modeling : Fable 5.1 trained a neural network producing a new elevation map of a third of the planet Venus, based on radar images from NASA’s Magellan mission. Resolution goes from 10 to 20 kilometers down to 2 to 3 kilometers, with heights up to 25% more accurate.
- Computational biology : Mythos 5.1 sped up seven open source deep learning models by up to 2.5 times on NVIDIA H100, by writing custom GPU kernels, which cuts the estimated GPU cost of genome-wide analyses by 30 to 60%.
Safety and safeguards
Three changes structure the safety side of this release.
Enterprise Frontier Safeguards
This new system gives enterprises the equivalent of a zero data retention policy : data is stored in cloud infrastructure controlled by the customer, not by Anthropic, with human review performed by default by the customer themselves. The rollout is phased starting this fall, on Claude Code, Claude Enterprise, the Claude Platform, Amazon Bedrock, the Google Cloud agent platform, and Microsoft Foundry.
More precise safeguards
In cybersecurity, Claude Code users can expect around 60% fewer interventions per session compared to Fable 5. Fable 5.1 can now be used to identify software vulnerabilities, but not to develop the corresponding exploits. Dual-use tasks, penetration testing, exploit generation, and binary-based vulnerability scanning, are still redirected to the Opus models. In biology, safeguards fire 85% less often for benign requests related to elementary biology and medical questions.
Anti-distillation mechanisms
Fable 5.1 hardens the protection against distillation. API accounts created from the launch date onwards can no longer manually edit Claude’s prior context in a multi-turn conversation while preserving the transcript of its thinking, a publicly documented technique for extracting the model’s reasoning.
info
Compliance with the EU AI Act
Anthropic is a signatory of the AI Act Code of Practice on transparency of AI-generated content. As such, an invisible watermark is added to the outputs of models released after August 2, 2026. It contains no information about the user, their organization, or their conversations, and has no effect on the quality of the responses. A detection API is in private preview for eligible organizations : regulators, law enforcement, media, fact-checkers, independent researchers, and educational organizations.
Availability
Claude Fable 5.1 is available immediately on all platforms : the Anthropic API under the identifier claude-fable-5-1, Claude.ai, Claude Code, Amazon Web Services, Google Cloud, and Microsoft Azure. Claude Mythos 5.1 remains restricted to organizations vetted through the CVP and LSVP programs. It also powers Claude Security, Anthropic’s code analysis product that scans for vulnerabilities and suggests patches for human review.
Conclusion
Fable 5.1 is less a break in capabilities than an answer to the adoption barriers of Fable 5. The performance gain is real, particularly on agentic scientific research where the score doubles, but it is the 75% reduction on cache reads that makes the model accessible to workloads previously reserved for Opus. For teams already in production on Fable 5, migration is straightforward, with one recommendation from Anthropic : take the opportunity to remove the defensive instructions added to compensate for the limits of the previous model.
Sources
Anthropic - Enterprise Frontier Safeguards
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.
