Anthropic has been building toward this moment for months.
In April, the company launched Claude Mythos Preview through Project Glasswing, a restricted release limited to cybersecurity professionals and critical infrastructure providers. At the time, Anthropic said it hoped to eventually make Mythos-level capabilities available to all users, but only once new safeguards were strong enough to prevent misuse.
Today, that day arrived.
Claude Fable 5 is Anthropic’s answer to the question of how you release an extremely powerful AI model to a general audience without creating serious risk.
The answer, it turns out, involves a layer of separate AI classifiers sitting on top of the model, ready to intercept dangerous requests and hand them off to a safer fallback.
It is the most significant Claude release since the original Mythos Preview, and it puts a Mythos-class model in the hands of anyone with a paid Claude subscription for the first time.
What Is Claude Fable 5?
Summary: Claude Fable 5 is a Mythos-class model, a tier of Claude models above Opus, made available for general use with safety classifiers layered on top.
Fable 5 is a Mythos-class model, a tier of Claude models that sit above Opus in capability.
The Mythos tier was introduced in April with Claude Mythos Preview, which was only available to a small group of vetted partners. Fable 5 is the same class of model made safe for general use, with a new set of classifiers handling the categories where Mythos-level capabilities pose the greatest risk.
The name itself is deliberate. Fable comes from the Latin fabula, meaning “that which is told,” akin to the Greek mythos.
The two models share the same underlying architecture. What separates them are the safeguards: Fable has them, while Mythos has some of them lifted for trusted partners.
According to Anthropic, Fable 5 is state-of-the-art on nearly every tested benchmark of AI capability. The longer and more complex the task, the larger its lead over previous Claude models.
That edge is most visible in four areas: software engineering, knowledge work, vision, and memory across long-context tasks.
What Fable 5 Can Do
Summary: Claude Fable 5 shows major gains in software engineering, knowledge work, vision, and long-context memory.
| Capability Area | What Anthropic Highlighted |
|---|---|
| Software engineering | Stripe reported that Fable 5 compressed months of engineering into days, including migrating a 50-million-line Ruby codebase in a single day, a task estimated at more than two months by hand. It also scored highest on Cognition’s FrontierCode evaluation. |
| Knowledge work | Fable 5 scored highest on Hebbia’s Finance Benchmark for senior-level reasoning and showed gains in document-based reasoning, chart and table interpretation, and problem solving. IMC reported that it aced their trading-analysis evaluations nearly across the board. |
| Vision | The model can extract precise numbers from detailed scientific figures, rebuild a web app’s source code from screenshots alone, and beat Pokémon FireRed using only raw game screenshots without maps or game-state data. |
| Memory and long-context | Fable 5 stays focused across millions of tokens in long-running tasks, improves its own outputs using persistent notes, and performed significantly better with file-based memory in Slay the Spire testing. |
The Safety Classifiers: How Anthropic Made It Safe for General Use
Summary: Anthropic uses separate AI classifiers to detect higher-risk requests and route those sessions to Claude Opus 4.8 instead of Fable 5.
Releasing a Mythos-class model to the general public comes with serious risks. Mythos-level capabilities in cybersecurity, biology, and chemistry could provide meaningful assistance to malicious actors in ways that existing models cannot.
Anthropic’s solution is a layer of separate AI classifiers that sit on top of Fable 5 and intercept requests in three categories before the main model ever responds.
When a request is flagged, the response is handled by Claude Opus 4.8 instead of Fable 5, and the user is informed that a fallback occurred.
Anthropic’s data shows that fewer than 5% of sessions trigger any fallback at all, meaning more than 95% of Fable 5 usage runs at full capability with no interruption.
Three categories trigger a fallback to Opus 4.8:
- Cybersecurity: Covers exploit development and agentic offensive tasks broadly.
- Biology and chemistry: Deliberately broad and likely to catch some harmless requests. Anthropic acknowledged this trade-off because the dual-use risk is significant enough to warrant caution.
- Distillation: Targets large-scale attempts to extract Claude’s capabilities to train competing models.
Anthropic ran extensive red-teaming against these classifiers, including an external bug bounty that produced no universal jailbreaks across more than 1,000 hours of testing.
One external partner found that Fable 5’s safeguards against harmful cyber queries were the most robust of any model tested. It complied with zero harmful single-turn requests across 30 different public jailbreak techniques.
Claude Mythos 5: The Version for Cyberdefenders
Summary: Claude Mythos 5 is the restricted version of the same underlying model, made available through Project Glasswing with some safeguards lifted for trusted partners.
Alongside Fable 5, Anthropic launched Claude Mythos 5 through Project Glasswing.
It is the same underlying model as Fable 5, but with some of the safety classifiers lifted. It is available today as an upgrade for all current Glasswing partners, including cybersecurity organizations and critical infrastructure providers who had access to Claude Mythos Preview.
Anthropic also announced plans to expand access to Mythos 5 through a broader trusted access program, and a separate trusted access program for biology researchers who need Fable 5 with the biology and chemistry classifiers removed.
A small number of biomedical researchers from fundamental and translational research organizations will be enrolled first, with plans to expand from there.
Pricing and Availability
Summary: Claude Fable 5 and Claude Mythos 5 cost $10 per million input tokens and $50 per million output tokens, with Fable 5 available through API and subscription plans during the initial rollout.
Both Fable 5 and Mythos 5 are priced at $10 per million input tokens and $50 per million output tokens, less than half the price of Claude Mythos Preview.
Developers can access Fable 5 via the Claude API today using the model string claude-fable-5. It is also fully available on consumption-based Enterprise plans from today.
For Pro, Max, Team, and seat-based Enterprise subscription plans, the rollout is staged:
- Now through June 22: Included at no extra cost.
- June 23: Removed from subscription plans, with usage credits required to access it.
- Later: Anthropic intends to restore Fable 5 as a standard subscription feature once capacity allows.
Mythos 5 remains restricted to Glasswing partners and, soon, a small group of biology researchers through the trusted access program.
What This Means for AI Search and Brand Visibility
Summary: A more capable Claude model can make brand recommendations more nuanced, which raises the bar for consistent entity data, structured content, reviews, and third-party mentions.
Fable 5 is a materially more capable model than anything Anthropic has previously made publicly available, which means the quality and specificity of Claude’s recommendations, including brand recommendations, will increase.
For marketers and brands focused on answer engine optimization, Claude is no longer a secondary platform. It is a primary one, and Fable 5 makes that more true.
The same signals that drive citation in ChatGPT and Gemini drive citation in Claude:
- Structured content that directly answers questions.
- Consistent entity data across directories and listing platforms.
- Multi-platform review presence.
- Authoritative third-party mentions.
What changes with a more capable model is its ability to synthesize across sources and make nuanced recommendations.
Brands with incomplete or inconsistent digital footprints will be more exposed, not less, as the underlying model gets better at identifying gaps.
For a full breakdown of how each AI platform decides which brands to cite, including Claude’s specific citation patterns, that resource covers the details.
Frequently Asked Questions
What is the difference between Claude Fable 5 and Claude Mythos 5?
They are the same underlying model. The difference is the safeguards. Fable 5 has classifiers that fall back to Opus 4.8 for requests related to cybersecurity, biology and chemistry, and distillation. Mythos 5 has some of those classifiers lifted and is only available to vetted partners through Project Glasswing and Anthropic’s trusted access program.
Is Claude Fable 5 available right now?
Yes. Fable 5 is available via the Claude API today using claude-fable-5. It is also included on Pro, Max, Team, and seat-based Enterprise plans at no extra cost through June 22, 2026. After that date, using it on subscription plans will require usage credits until Anthropic has sufficient capacity to restore it as a standard feature.
How much does Claude Fable 5 cost?
Claude Fable 5 costs $10 per million input tokens and $50 per million output tokens. This is less than half the price of Claude Mythos Preview, which it effectively replaces for general-use cases.
What happens when Fable 5’s safety classifiers are triggered?
The request is handled by Claude Opus 4.8 instead of Fable 5, and the user is notified that a fallback occurred. Anthropic’s data shows this happens in fewer than 5% of sessions. Opus 4.8 is a highly capable model in its own right, so a fallback response is still useful. It is not a refusal.
What is Project Glasswing?
Project Glasswing is Anthropic’s program for deploying Mythos-class models to a vetted group of cyberdefenders and critical infrastructure providers in collaboration with the US government. It launched in April 2026 with Claude Mythos Preview and is now being upgraded to Claude Mythos 5.
Does Claude Fable 5 change how Claude cites brands in AI search?
The underlying citation signals remain the same: structured content, consistent entity data, reviews across multiple platforms, and authoritative third-party mentions. What changes is the model’s ability to synthesize across sources and make more nuanced recommendations. A more capable model is better at identifying gaps in a brand’s digital presence, which makes AEO readiness more important, not less.
