Anthropic releases its first Mythos-class model Claude Fable

Overview

Anthropic released Claude Fable 5 on June 9, 2026, calling it the most capable model it has ever made widely available. Fable 5 is the first public release from the company's new Mythos class, a tier the company had held back over fears it was too dangerous to share. Anthropic says the public launch became possible because of new safeguards that block responses in cybersecurity, biology, chemistry, and a few other high-risk areas, sending those flagged queries to the older Opus 4.8 model instead. A restricted, less-guarded sibling called Mythos 5 is reserved for vetted cybersecurity and government partners.

Key Takeaways

Claude Fable 5 is the first broad release from Anthropic's Mythos class, a model family the company previously said was too dangerous to publish because of its cybersecurity strength.
Safeguards reroute flagged requests about cybersecurity, biology, chemistry, and model distillation to the older Claude Opus 4.8, and Anthropic says this happens in fewer than 5% of sessions.
On SWE-Bench Pro, a real-world software engineering test, Fable 5 leads Opus 4.8 by about eleven points, and roughly doubles it on the hardest agentic coding and reasoning tasks.
Mythos 5 is the same underlying model with safeguards removed, offered only to vetted cybersecurity professionals and government partners through Project Glasswing.
Fable 5 is available immediately through the Claude API, Claude Code, enterprise plans, and the AWS, Google Cloud, and Microsoft Foundry clouds.
Anthropic now keeps a mandatory 30-day data retention policy on all Mythos-class traffic for safety monitoring, with the data not used to train models.

Stats & Key Facts

#Pricing is set at $10 per million input tokens and $50 per million output tokens, double the cost of Opus 4.8.
#Safeguards trigger and reroute queries in fewer than 5% of sessions, leaving more than 95% running on the full Mythos-class model.
#Fable 5 opens about an eleven-point lead over Opus 4.8 on the SWE-Bench Pro software engineering benchmark.
#An external bug bounty ran more than 1,000 hours of testing and produced no universal jailbreaks.
#In life sciences testing, Fable 5 helped accelerate drug design roughly tenfold and produced strong candidates for 9 of 14 protein targets.
#Anthropic enforces a 30-day data retention window on all Mythos-class traffic for safety review.

Anthropic releases its first Mythos-class model Claude Fable

What the Mythos Class Means for Anthropic's Model Lineup

Mythos sits above Anthropic's existing Opus class as a new top tier.

Anthropic groups its models into capability tiers, and Mythos is the newest and most capable group. The company had described this family as so strong at cybersecurity work that releasing it openly carried real risk, so it held the models back from the general public.

Fable 5 is the version of Mythos that Anthropic considers safe enough for general use. It carries the same underlying intelligence as the restricted Mythos 5 model, but with extra guardrails layered on top. The company frames this split as a way to share most of the model's value while keeping the most sensitive abilities locked down.

How the Cybersecurity and Biology Safeguards Reroute Risky Queries

The safety system works by detecting sensitive requests and handing them to a safer model.

›Classifiers watch for requests tied to cybersecurity exploitation, biology and chemistry misuse, and model distillation, which is copying a model's abilities to train a rival.
›When a request trips one of these classifiers, Fable 5 falls back to the older Claude Opus 4.8 instead of answering with full Mythos-class power.
›Anthropic reports this fallback happens in fewer than 5% of sessions, so more than 95% of the time users get the full model.
›An external bug bounty logged over 1,000 hours of testing and found no universal jailbreaks, and outside red-team groups reported the same on long agentic tasks.

Fable 5 Benchmark Results Against Opus 4.8

Anthropic reports state-of-the-art scores across most tests it ran.

›On SWE-Bench Pro, a benchmark built around real software engineering work, Fable 5 leads Opus 4.8 by roughly eleven points.
›Its largest margins show up on the hardest agentic and frontier-reasoning tasks, where it nearly doubles Opus 4.8 on tests named FrontierCode and Blueprint-Bench.
›Anthropic claims new top scores in vision tasks and on a senior-level finance reasoning benchmark from Hebbia.
›The company says the model's lead over rivals widens as tasks grow longer and more complex.

Real-World Coding and Science Results Anthropic Cited

Anthropic pointed to early partner results to show practical impact.

On the software side, Anthropic said the payments company Stripe used the model to compress months of work into days, including a migration of a 50-million-line Ruby codebase that took one day instead of an estimated two months by hand. Coding tool makers including GitHub and Cursor described the model as better at long, multi-step tasks.

In life sciences, Anthropic said internal protein-design experts saw drug design accelerate about tenfold, with strong candidates emerging for 9 of 14 protein targets. Scientists reportedly preferred the model's novel molecular biology hypotheses around 80% of the time. The company also said persistent memory let the model perform about three times better than Opus 4.8 at the game Slay the Spire across long runs.

Mythos 5, Project Glasswing, and Restricted Access

The unguarded version stays behind controlled access programs.

Mythos 5 is the same model as Fable 5 with the safeguards lifted, and Anthropic says it has the strongest cybersecurity abilities of any model in the world. Because of that, the company restricts it to vetted users rather than the open public.

Initial access runs through Project Glasswing, a collaboration tied to US government work, where partners get the cybersecurity safeguards removed. Anthropic also described planned trusted-access programs for biology researchers and for cybersecurity organizations, each lifting only the safeguards relevant to that group while keeping the others in place.

Where Fable 5 Is Available and What It Costs

The model shipped across major platforms on launch day.

›Fable 5 is available through the Claude API, Claude Code, and consumption-based enterprise plans.
›It also runs on Amazon Web Services, Google Cloud, and Microsoft Foundry.
›Pricing is $10 per million input tokens and $50 per million output tokens, twice the rate of Opus 4.8.
›Anthropic applies a mandatory 30-day data retention policy to all Mythos-class traffic, used for safety review rather than training, with logged human access and deletion after the window in nearly all cases.

Why the Public Release Followed a Danger Warning

The launch came days after Anthropic flagged rising AI risk.

Reporting on the release noted that Anthropic published it shortly after warning that frontier AI was advancing fast enough to raise safety concerns, including risks tied to systems that improve themselves with little human oversight. The Mythos family was held back at first for exactly these reasons.

The company's stance is that the new safeguard system is what changed the calculation. By blocking the most sensitive request types and routing them to a safer model, Anthropic argues it can share most of Fable 5's value without handing out the abilities that made the raw Mythos models too risky to publish.

Frequently Asked Questions

What is Claude Fable 5?

Claude Fable 5 is Anthropic's most capable widely available AI model, released on June 9, 2026. It is the first public model from the company's new Mythos class and comes with built-in safeguards for high-risk topics.

How is Fable 5 different from Mythos 5?

They share the same underlying model, but Fable 5 keeps safeguards active while Mythos 5 has them removed. Mythos 5 is restricted to vetted cybersecurity professionals and government partners, while Fable 5 is available to the public.

What do the safeguards actually block?

Classifiers detect requests about cybersecurity exploitation, biology, chemistry, and model distillation, then reroute those answers to the older Opus 4.8 model. Anthropic says this happens in fewer than 5% of sessions.

How much does Claude Fable 5 cost to use?

API pricing is $10 per million input tokens and $50 per million output tokens, which is double the rate of Claude Opus 4.8.

Where can businesses access Fable 5?

It is available through the Claude API, Claude Code, consumption-based enterprise plans, and the AWS, Google Cloud, and Microsoft Foundry cloud platforms.

Claude Fable 5 marks the first time Anthropic has opened its Mythos tier to the public, betting that automated safeguards can share the model's strengths while holding back its most sensitive abilities. The fully unguarded Mythos 5 stays locked behind vetted partner programs.