Anthropic has introduced Claude Fable 5.1 and Claude Mythos 5.1, based on the same underlying model with different safeguard levels. Fable 5.1 is for general use, while Mythos 5.1 is aimed at vetted cybersecurity and life sciences users.
Claude Fable 5.1
Claude Fable 5.1 is designed for coding, knowledge work and long-running problem-solving. Anthropic says Low and Medium effort can match or exceed Fable 5 on some workloads while using less compute.
It defaults to High effort in Claude Code and Medium effort in Claude Cowork and Claude.ai. Investment firm Millennium also used it to identify the cause of a rare internal system crash that had remained unexplained for several years.
Scientific research
Anthropic tested the models across molecular design, computational analysis and computational biology.
Mythos 5.1 used open-source protein design and folding tools for molecular design. Results included:
- 10x higher binding affinity than the best Adaptyv Bio competition designs on three targets.
- Nearly 50% hit rate for viable binders across 12 targets.
- Typical protein design hit rate of 10% to 15%.
Fable 5.1 created a high-resolution elevation map covering about one-third of Venus using NASA Magellan radar images. It provides 2–3 km detail instead of 10–20 km, with heights measured up to 25% more accurately. Anthropic plans to release it under a Creative Commons license ahead of NASA’s VERITAS and ESA’s EnVision missions.
Mythos 5.1 also created custom GPU kernels and cached intermediate results to accelerate seven open-source deep learning models by up to 2.5x, with identical outputs. The optimizations could reduce GPU costs by 30% to 60% and will be open-sourced.
Scientific support includes:
- Model Hardware Standard: Claude can operate laboratory equipment directly and safely.
- AI for Science: Free credits for researchers working on high-impact projects.
- Claude Team: Discounted usage for scientists.
Safety and alignment
Anthropic evaluated the models for chemical and biological risks, cybersecurity, agentic safety and alignment through automated evaluations, expert red-teaming and external testing.
Mythos 5.1’s chemical and biological capabilities are higher than Mythos 5 but remain below the next risk tier under the Responsible Scaling Policy. Cybersecurity testing with safeguards disabled found its highest measured cyber capability among Anthropic’s released models, while remaining in the lower risk category of the Frontier Compliance Framework.
Fable 5.1’s cyber safeguards were tested internally, by two external organizations and by Gray Swan. No critical-severity jailbreak was found, consistent with Fable 5 and Opus 5.
Mythos 5.1 refused malicious agentic coding and computer-use requests at rates comparable to Mythos 5, Sonnet 5 and Opus 5. Alignment testing covered behavioral evaluations, natural-language autoencoder analysis, misalignment tests, training-data reviews, internal use and external testing.
Compared with Mythos 5, it was less likely to access resources outside its test environment, use motivated reasoning or ignore explicit constraints. Training-data reviews also found lower rates of attempted and successful reward hacking, although it can still bypass approvals and auto-mode classifiers.
Safeguards and anti-distillation
Enterprise Frontier Safeguards (EFS) provide privacy equivalent to zero data retention while allowing misuse detection. Customer data remains in customer-controlled cloud infrastructure, with human review handled by the customer by default.
EFS was developed with more than 100 customers across financial services, healthcare, manufacturing, telecom, law, retail and the public sector, along with AWS, Google Cloud and Microsoft Azure.
It supports:
- Claude Code
- Claude Enterprise
- Claude Platform
- Amazon Bedrock
- Claude Platform on AWS
- Google’s Agent Platform
- Microsoft Foundry
Fable 5.1’s biology safeguards trigger 85% less often for benign elementary biology and medical questions than those introduced with Fable 5. It can identify software vulnerabilities for defensive work, while penetration testing, exploit generation and binary-based vulnerability scanning remain redirected to Opus models. Life sciences R&D queries also continue to be directed to Opus models.
Claude Code users see around 60% fewer cyber safeguard interventions per session than with previous Fable 5 safeguards.
For anti-distillation, new API accounts created from today onwards cannot manually edit Claude’s prior context in multi-turn conversations while retaining the transcript of Claude’s prior thinking. The change is rolling out gradually, and existing accounts are not currently affected.
Trusted access and EU AI Act
Mythos 5.1 is offered through:
- Cyber Verification Program (CVP): Reduced cyber safeguards for defensive security work, with Mythos-class models planned for inclusion.
- Life Sciences Verification Program (LSVP): Mythos 5.1 access for professional life sciences R&D while retaining its other safeguards.
LSVP was developed with the US government, with the first participants enrolled. Claude Security, which scans codebases for vulnerabilities and suggests patches for human review, is now powered by Mythos 5.1.
In July 2026, Anthropic and 190 other signatories signed the EU AI Act’s Code of Practice on Transparency of AI-Generated Content. Models released after August 2, 2026 must include a watermark estimating the likelihood Claude helped write text.
The watermark is invisible without the detection API, has no practical impact on output quality or content, and contains no user, organization or conversation information. The API is in private preview for regulators, law enforcement, media, fact-checkers, independent researchers, educational organizations, EU civil society groups and enterprises with compliance requirements.
Pricing and availability
Claude Fable 5.1 is available today across Anthropic’s platforms, including AWS, Google Cloud and Microsoft Azure. Developers can use claude-fable-5-1 through the Claude API.
Cache-read pricing is 75% lower at $0.25 per million tokens. Typical workloads cost around 25% less than Fable 5, while complex coding and highly agentic workloads can save up to approximately 45%.
Fable 5.1 pricing:
- Input: $10 per million tokens
- Output: $50 per million tokens
Claude Mythos 5.1 is available to vetted cyberdefenders and life scientists through the trusted access programs. It is currently limited to a set of US organizations, with expansion to additional domestic and international partners planned.