Claude Fable 5 is the most powerful artificial intelligence model Anthropic has ever made generally available. Launched on June 9, 2026, it is the first Mythos-class model, capable of running complex, long-horizon tasks for several days, with state-of-the-art performance on nearly every benchmark in software engineering, vision and knowledge work.
In this complete guide, we break down everything you need to know about Claude Fable 5: its real capabilities, pricing, security safeguards, its mysterious twin Mythos 5, and the concrete use cases that are changing the game for developers and businesses.
What is Claude Fable 5?
Claude Fable 5 is a generative AI model developed by Anthropic, the company founded by Dario Amodei and several former OpenAI members. It inaugurates the fifth generation of Claude models and belongs to the "Mythos" family, until now reserved for restricted use because of capabilities deemed too sensitive.
The big news? Anthropic has made a Mythos-class model accessible to the public, but wrapped it in strong safeguards. In practice, Fable 5 is the same engine as Mythos 5, but restricted in certain high-risk areas such as cybersecurity and biology.
The term "Fable" refers to the secured version for general use, while "Mythos" refers to the unfiltered version reserved for trusted partners. Both share the exact same underlying model.
Why the number "5"?
This is one of the launch's mysteries: why number these models "5" when no version 1 through 4 of Fable or Mythos was ever released publicly? Anthropic has not officially commented on this choice. The most likely hypothesis is an alignment with the internal naming of model generations, with Fable 5 conceptually succeeding the Opus 4.x lineage.
Availability and access
Fable 5 is available natively on the Claude Platform, via the API (claude-fable-5), as well as on Amazon Web Services, Google Cloud and Microsoft Foundry. For organizations, it is offered on the consumption-based Enterprise plan.
For individuals, Anthropic adopted a gradual approach: from June 9 to 22, 2026, Fable 5 was accessible at no extra cost in paid subscriptions (Pro, Max, Team). After this test phase, it temporarily switches to a consumption-based credit system, before a permanent reintegration planned later in the year, once network infrastructure is strengthened.
Claude Fable 5's capabilities
Anthropic describes Fable 5 as its most capable model to date, with one striking feature: the longer and more complex a task, the larger its lead over competing models. Let's review its areas of strength.
Software engineering
This is undoubtedly where Fable 5 impresses most. During early testing, payment platform Stripe reported that the model compressed months of engineering into a few days. The most striking example: the complete migration of a 50-million-line Ruby codebase, completed in a single day, where an entire team would have taken over two months by hand.
Fable 5 is also more token-efficient than its predecessors. On Cognition's FrontierCode evaluation, which tests the ability to produce production-quality code, it scores highest among frontier models, even at medium effort.
In an agent harness like Claude Code, Fable 5 can work for days at a time: multi-stage planning, delegating to sub-agents, and checking its own work.
Vision
Fable 5 sets a new state of the art for vision tasks. It can extract precise numbers from detailed scientific figures, or rebuild a web app's source code from screenshots alone.
The most emblematic example remains Pokémon, which has become a sort of unofficial benchmark for AI. Fable 5 completed the game Pokémon FireRed entirely by analyzing only raw screenshots, with no maps or navigation aids. Previous versions of Claude failed even with dedicated assistance tools.
Knowledge work and analysis
On complex analytical tasks, Fable 5 also shines. On Hebbia's Finance benchmark, designed for senior-level reasoning, it achieves the highest score ever recorded, with notable improvements in interpreting charts, tables and documents.
It is also the first model to break the 90% mark on certain long-running analytics benchmarks, a 10-point jump over Opus. Law firms report that its contract redlines match or surpass previous models in blind tests.
Memory and long context
Fable 5 stays focused across millions of tokens in long-running tasks and improves its outputs using its own notes. In a test on the deck-building game Slay the Spire, access to persistent file-based memory improved its performance three times more than for Opus 4.8.
Fable 5's security safeguards
Making such a powerful model accessible to the public comes with risks. Without safeguards, Fable 5's capabilities in cybersecurity or biology could be misused to cause serious damage. Anthropic therefore deployed an unprecedented protection system.
The fallback mechanism to Opus 4.8
The principle is simple: when Fable 5's classifiers detect a sensitive request, the response is automatically handled by Claude Opus 4.8, a less capable but still high-performing model. The user is informed each time this fallback occurs.
Anthropic deliberately tuned these safeguards conservatively. They can therefore sometimes block harmless requests, but on average they only trigger in less than 5% of sessions.
The three areas covered
Fable 5's classifiers cover three main categories of high-risk requests:
| Area | Identified risk |
|---|---|
| Cybersecurity | Discovery and exploitation of vulnerabilities, agentic hacking (reconnaissance, lateral movement) |
| Biology and chemistry | Uplift for high-risk biological research, design of pathogens |
| Distillation | Attempts to extract the model's capabilities to train rival AIs |
On cybersecurity, Anthropic conducted intensive red-teaming. An external bug bounty of over 1,000 hours produced no universal jailbreak. One partner even found that Fable 5's safeguards were the most robust of all models tested, including Opus 4.8 and 4.7.
A new data retention policy
Anthropic now requires 30-day data retention for all traffic on Mythos-class models, including for business customers. This data is not used to train new models, but only for security purposes: detecting new attacks and reducing false positives. Human access is logged and data is deleted after 30 days.
Claude Mythos 5: the unfiltered version
Alongside Fable 5 intended for the general public, Anthropic launched Claude Mythos 5. It is the exact same underlying engine, but with safeguards lifted in certain areas, notably cybersecurity.
This unrestricted version is not accessible to everyone. Access remains strictly reserved for organizations that are members of Project Glasswing, an initiative led in collaboration with the US government. The program is currently open to organizations from more than fifteen countries.
Mythos 5 has the strongest cyberdefense capabilities of any model in the world. It is deployed to help defenders secure critically important infrastructure software.
Spectacular scientific results
Mythos 5 demonstrated remarkable scientific capabilities. In protein design, Anthropic's internal experts accelerated certain steps of the drug discovery process by around tenfold. Of 14 protein targets studied, nine yielded promising candidates for treatment development.
In molecular biology, Mythos 5 is the first model to consistently produce novel, compelling scientific hypotheses. In blind comparisons, scientists preferred its hypotheses about 80% of the time. One of them, on a new mechanism of an E. coli protein, was even corroborated by a lab working independently on the same subject.
In genomics, the model conducted over a week of autonomous research, assembling single-cell data from millions of cells spanning 138 animal species, and designed a machine learning model that outperformed a model recently published in the journal Science, while being 100 times smaller.
Claude Fable 5 pricing
This power boost comes at a cost. On the API, Fable 5 is billed at:
| Token type | Price per million |
|---|---|
| Input tokens | $10 |
| Output tokens | $50 |
This is exactly double the rates of Claude Opus 4.8, but represents a 50% reduction compared to the Mythos Preview evaluation version. The model retains the 90% discount on input tokens through prompt caching.
For workloads that need to stay in the US, US-only inference is available at 1.1x the standard rate, for both input and output. Good news: you are never charged Fable prices for requests rerouted to Opus 4.8.
When should you use Claude Fable 5?
Anthropic offers a range of models covering the full spectrum of speed / price / performance. Fable 5 is the most capable model available to the public, but it isn't necessarily the right choice for every task. Here are the use cases where it truly makes a difference:
Autonomous agents
In a harness like Claude Code or Claude Managed Agents, Fable 5 can work for entire days: planning stages, delegating to sub-agents and checking its own work. This is ideal for complex multi-agent workflows.
Ambitious coding projects
Large migrations, complex implementations, multi-day autonomous sessions: Fable 5 excels. It writes its own tests to verify its work and uses vision to compare its results against the original goals.
Enterprise workflows
Fable 5 handles multi-stage knowledge work with minimal supervision, from deep research to deliverables ready for review. Teams can hand off large projects and review completed work rather than supervising every step.
Vision-heavy tasks
For document-rich domains (finance, legal, analytics, architecture), Fable 5 understands diagrams, charts and tables nested in files and PDFs.
Strengths and limitations of Claude Fable 5
To help you decide, here is an objective summary of the model's strengths and constraints.
| Strengths | Limitations |
|---|---|
| State of the art on most benchmarks | High price (double that of Opus 4.8) |
| Long-horizon tasks (several days) | Sometimes overly strict safeguards (false positives) |
| Advanced vision and reasoning | Mandatory 30-day data retention |
| More token-efficient than previous models | Mythos capabilities reserved for partners |
Frequently asked questions
What is the difference between Claude Fable 5 and Mythos 5?
Both share the exact same underlying model. Fable 5 is the secured version for the general public, with safeguards that fall back to Opus 4.8 on sensitive topics. Mythos 5 is the unfiltered version, reserved for a restricted circle of trusted partners via Project Glasswing.
How much does Claude Fable 5 cost?
On the API, Fable 5 costs $10 per million input tokens and $50 per million output tokens, double that of Opus 4.8. A 90% discount applies to input tokens via prompt caching. You are not charged Fable prices for requests rerouted to Opus 4.8.
Is Claude Fable 5 really dangerous?
The model has advanced cybersecurity and biology capabilities that could be misused. That is why Anthropic deployed safeguards blocking or limiting these areas. For typical use, more than 95% of sessions run entirely on Fable 5 with no fallback at all.
How do you access Claude Fable 5?
Fable 5 is available via the API (claude-fable-5), on the Claude Platform, as well as on AWS, Google Cloud and Microsoft Foundry. For organizations, it is accessible on the Enterprise plan. Individuals can access it via paid subscriptions during test periods, then via a credit system.
What does "Mythos class" mean at Anthropic?
The Mythos class refers to Anthropic's most advanced range of models, until now deemed too powerful to deploy widely without safeguards. Fable 5 is the first Mythos model made accessible to the general public, thanks to new security protections.