Anthropic has launched Claude Fable 5.1 and Claude Mythos 5.1, two versions of the same underlying frontier model designed to push AI further into coding, knowledge work, scientific research and cybersecurity.
The company describes the models as its most advanced systems for coding and knowledge work. Fable 5.1 is generally available, while Mythos 5.1 is restricted to vetted organizations working in areas such as cybersecurity and life sciences because of its more permissive safeguards.
The launch is notable not only because of benchmark improvements. Anthropic is also targeting three practical problems that have become increasingly important as AI agents become more capable: cost, safety and privacy.
Fable 5.1 gets a 75% reduction in cache-read pricing, while Anthropic says typical workloads can cost about 25% less and highly agentic workloads can be up to approximately 45% cheaper than Fable 5.
What Are Claude Fable 5.1 and Mythos 5.1?
Fable 5.1 and Mythos 5.1 are not completely separate base models.
According to Anthropic, both use the same underlying model, but they are deployed with different levels of safeguards.
Claude Fable 5.1 is the broadly available version intended for coding, professional knowledge work, research and long-running agentic tasks.
Claude Mythos 5.1 provides more permissive safeguards for vetted organizations whose work requires advanced cybersecurity or life-sciences capabilities. Access is currently limited through Anthropic’s trusted-access programs.
This approach allows Anthropic to expose the same underlying intelligence to different users while applying stronger restrictions where the potential misuse risk is higher.
Claude Fable 5.1 Brings a Major Coding Upgrade
Coding is one of the biggest areas of improvement in Fable 5.1.
Anthropic reports a 55.8% score on Terminal-Bench 4.0, compared with 42.0% for Fable 5 in its evaluation setup. Mythos 5.1 reached 60.9% on the same benchmark.
On CursorBench 3.2.0, Fable 5.1 reached 73.4% at maximum effort, compared with 70.5% for Fable 5.
The improvements are aimed at more than simply generating code snippets.
Anthropic says Fable 5.1 is better at:
- Understanding large and complicated codebases
- Finding the root cause of software problems
- Verifying its own work
- Handling long-running coding tasks
- Working through multiple steps without constant human intervention
- Building prototypes and completing complex software workflows
Anthropic also highlighted testing from companies including Red Hat, Datadog, MongoDB, Plaid and others. These are customer or partner evaluations rather than independent universal benchmarks, so they should be viewed as supporting evidence rather than proof that Fable 5.1 is better at every coding task.
One particularly interesting example came from Millennium, where Anthropic says Fable 5.1 traced an extremely rare crash to a bug in an external vendor library after previous attempts had failed to identify the cause.
Scientific Research Is Another Major Focus
Anthropic is positioning the new models as research assistants capable of doing more than summarizing scientific literature.
In one experiment, Fable 5.1 created a high-resolution elevation map covering approximately one-third of Venus using radar imagery from NASA’s Magellan mission and existing mapping data.
Anthropic says the resulting map can reveal geological details at roughly two- to three-kilometer resolution, compared with approximately 10- to 20-kilometer resolution in the previous data used for comparison.
Mythos 5.1 was also tested on protein-design tasks.
Anthropic says its model-generated protein binders were experimentally validated by external organizations. Across 12 targets, the model achieved a hit rate of nearly 50%, while typical protein-design hit rates are around 10% to 15%, according to Anthropic. On three targets, some designs reportedly achieved binding affinities 10 times higher than the best designs submitted to Adaptyv Bio’s protein-design competitions.
These results are particularly important because they illustrate a shift from AI being primarily a text-generation tool toward being an agent that can combine reasoning, tools, computation and experimentation.
Mythos 5.1 Pushes Further Into Cybersecurity
Cybersecurity is where the distinction between Fable and Mythos becomes especially important.
Anthropic says Mythos 5.1 demonstrated the strongest cybersecurity capabilities of any model it has released, although the company says its evaluations still place it within the lower risk category of its Frontier Compliance Framework.
For Fable 5.1, Anthropic has simultaneously made its cybersecurity safeguards more precise.
The model can now be used to identify software vulnerabilities as part of defensive security work. However, Anthropic says its safeguards still redirect or restrict higher-risk activities such as:
- Penetration testing
- Exploit generation
- Binary-based vulnerability scanning
Anthropic reports that Claude Code users can expect approximately 60% fewer cybersecurity-safeguard interventions per session compared with the previous Fable 5 safeguards.
That change matters because overly aggressive AI safety systems can sometimes interfere with legitimate defensive cybersecurity work.
The goal is therefore not simply to make safeguards weaker, but to make them more precise.
Anthropic Also Improved Biology Safeguards
Anthropic has made similar changes in biology.
The company says its newer biology safeguards for Fable 5.1 and Fable 5 trigger 85% less often on benign elementary-biology and medical questions compared with the safeguards originally launched with Fable 5.
At the same time, research and development requests involving life sciences remain subject to stronger restrictions and may be routed to other models or controlled access programs.
Mythos 5.1 is being made available to selected life-science researchers through a trusted-access program.
This reflects Anthropic’s broader approach of separating legitimate research use from higher-risk biological capabilities.
Fable 5.1 Is Also Cheaper to Run
One of the less flashy but potentially most important changes is pricing.
Anthropic has reduced cache-read pricing by 75%, bringing it to $0.25 per million tokens.
The rest of the token pricing remains:
- Input: $10 per million tokens
- Output: $50 per million tokens
- Cache reads: $0.25 per million tokens
Anthropic estimates that the pricing change can reduce typical Fable 5.1 workloads by around 25%, while highly agentic and context-heavy workloads can see savings of up to approximately 45%.
This is especially relevant for AI agents.
Traditional chatbot interactions may only require a handful of requests. Autonomous coding and research agents can repeatedly read large amounts of context, call tools and revisit previous information.
Reducing the cost of cached context can therefore have a much larger impact on long-running agent workflows than a simple reduction in the headline input-token price.
Fable 5.1 Gets Better at Long-Running Agent Work
Anthropic is increasingly designing Claude around the idea that AI models will operate for extended periods rather than simply answer individual prompts.
Fable 5.1 was evaluated on agentic scientific research, terminal coding, multidisciplinary reasoning, business workflows, computer use and agentic coding.
The company reports a 31.4 score on AutomationBench, compared with 17.1 for Fable 5. It also reports 60.9% on Humanity’s Last Exam without tools and 65.0% with tools.
The more important shift, however, is behavioral.
Anthropic says Fable 5.1 is better at avoiding shortcuts, identifying root causes and verifying its own work. That is particularly valuable for agents because an AI that produces an apparently correct answer but fails to validate it can create expensive downstream problems.
New Enterprise Privacy Features Are Coming
Anthropic is also addressing enterprise concerns around data retention.
The company announced Enterprise Frontier Safeguards, or EFS, which are designed to let customers keep their data in cloud infrastructure controlled by the customer rather than Anthropic.
Anthropic describes the resulting setup as providing privacy comparable to a zero-data-retention arrangement while still supporting mechanisms for detecting and responding to misuse. The feature is scheduled to roll out to enterprise customers in phases beginning later in fall 2026.
EFS is planned for Claude Code, Claude Enterprise, the Claude Platform and selected cloud environments including Amazon Web Services, Google Cloud and Microsoft Azure.
For companies considering autonomous AI agents, this could become just as important as raw model intelligence.
Anthropic Is Also Fighting AI Model Distillation
Fable 5.1 introduces stronger anti-distillation mechanisms.
Model distillation involves extracting capabilities from a more capable AI model and transferring them into another model. Anthropic argues that large-scale distillation can become a safety problem when advanced capabilities are extracted and subsequently deployed without comparable safeguards.
The company has therefore changed how new API accounts can manipulate prior conversational context while retaining Claude’s previous reasoning context, closing a technique Anthropic says could be used to illicitly extract model capabilities.
Existing accounts are not immediately affected by the change, according to Anthropic.
How Fable 5.1 Compares With Other Frontier Models
Anthropic’s own benchmark table places Fable 5.1 ahead of its listed predecessors and several competing models on a number of evaluations.
For example, Anthropic reports:
| Benchmark | Claude 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 |
| CursorBench 3.2.0 | 73.4% | 70.5% | 70.0% | 67.2% |
Anthropic’s benchmark methodology and evaluation setup should be kept in mind when interpreting these numbers. Some tests have relatively large statistical uncertainty, and benchmark results do not necessarily translate directly into better performance for every real-world workload.
Still, the overall direction is clear: Anthropic is trying to make Fable a model that can work through complex tasks for longer, use tools more effectively and produce useful results at a lower operating cost.
Read More:- OpenAI Publishes GPT-6 Astra Safety Review as Model Reaches Critical Cybersecurity Level
Who Can Use Claude Fable 5.1?
Fable 5.1 is generally available through Anthropic’s ecosystem and cloud partners.
Developers can access the model through the Claude API using the model identifier claude-fable-5-1. Anthropic also lists availability through Amazon Web Services, Google Cloud and Microsoft Azure.
Mythos 5.1 is different.
Access is restricted to vetted organizations through trusted-access programs focused on cybersecurity and life sciences. Anthropic says it is gradually expanding these programs rather than making Mythos broadly available.
Why the Claude 5.1 Launch Matters
The biggest story here isn’t simply that Anthropic released a model with higher benchmark scores.
It is the combination of intelligence, autonomy, economics and safety.
Fable 5.1 is designed to handle longer coding and research workflows while costing less for workloads that heavily reuse context.
Mythos 5.1 demonstrates how the same underlying model can be deployed with different safety boundaries when capabilities become powerful enough to create meaningful cybersecurity or biological risks.
And Anthropic is simultaneously trying to reduce false-positive safety interventions, improve enterprise privacy and prevent unauthorized capability extraction.
That suggests the next phase of the AI race may be less about producing the biggest benchmark number and more about answering a difficult question:
Can frontier AI become capable enough to perform valuable work autonomously without becoming too expensive, unreliable or dangerous to deploy?
What Happens Next for Claude?
Anthropic says it expects to continue expanding access to Mythos through its trusted programs while rolling out Enterprise Frontier Safeguards later this year.
The company also plans to open-source some of the computational optimizations produced during its scientific research work.
For developers, Fable 5.1’s lower cache-read pricing could make long-running coding and agentic applications more economically viable.
For researchers, the model’s demonstrated ability to combine literature, code, computation and external tools points toward increasingly autonomous scientific workflows.
But the real test will come outside Anthropic’s controlled evaluations.
If Fable 5.1 can consistently maintain accuracy and reliability across long-running production tasks while delivering the promised cost improvements, it could become one of the more practical frontier models for enterprise AI agents.
Final Takeaway
Claude Fable 5.1 is more than a routine model refresh.
Anthropic has combined significant coding and research improvements with lower cache costs, more precise cybersecurity and biology safeguards, stronger enterprise privacy mechanisms and new anti-distillation protections.
Mythos 5.1 goes further in cybersecurity and life-sciences capabilities but remains restricted to vetted organizations.
The broader significance is clear: frontier AI development is moving from better answers toward AI systems capable of completing entire multi-step projects.
The companies that can make those systems powerful, affordable, reliable and safe enough for real-world deployment will likely have the biggest influence on the next stage of the AI industry.
FAQ
What is Claude Fable 5.1?
Claude Fable 5.1 is Anthropic’s latest generally available frontier AI model focused on coding, knowledge work, research and long-running agentic tasks.
What is Claude Mythos 5.1?
Claude Mythos 5.1 is the same underlying model as Fable 5.1 but with more permissive safeguards for vetted organizations working in cybersecurity and life sciences.
Is Claude Fable 5.1 better than Fable 5?
Anthropic reports substantial improvements across coding, scientific research, knowledge work, computer use and agentic tasks. For example, it reports 55.8% on Terminal-Bench 4.0 versus 42.0% for Fable 5 in its evaluation setup.
How much does Claude Fable 5.1 cost?
Anthropic lists pricing of $10 per million input tokens, $50 per million output tokens and $0.25 per million cache-read tokens. Cache-read pricing is 75% lower than before.
Is Claude Mythos 5.1 publicly available?
No. Mythos 5.1 is currently limited to vetted organizations through Anthropic’s trusted-access programs.
Can Claude Fable 5.1 find software vulnerabilities?
Yes. Anthropic says Fable 5.1 can be used for defensive vulnerability identification. However, safeguards still restrict or redirect higher-risk activities such as exploit generation and penetration testing.
Is Claude Fable 5.1 cheaper than Fable 5?
Anthropic estimates typical workloads can cost around 25% less, while highly agentic workloads can see savings of up to approximately 45%, primarily because cache-read pricing has been reduced.
