Anthropic Releases Claude Fable 5: The Most Powerful AI Model Ever Made Publicly Available — and It Comes With a Catch

June 9, 2026 — Anthropic did something it swore it would never do: it handed the general public a model from its top-secret Mythos tier. For months, that capability class had been locked behind government agreements and classified cybersecurity partnerships. Now it has a new name — Claude Fable 5 — and anyone with a Pro subscription can use it. But the story doesn't end there.

What Is Claude Fable 5, and Why Does It Matter?

To understand the significance of this launch, you need a little backstory.

Back in April 2026, Anthropic unveiled Claude Mythos Preview — a model so capable at finding and exploiting software vulnerabilities that the company publicly refused to release it to the public. Anthropic said, bluntly, that AI had crossed a threshold: models could now "surpass all but the most skilled humans at finding and exploiting software vulnerabilities." Access was restricted to roughly 50 vetted cybersecurity partners through a program called Project Glasswing, which included major organizations like AWS, Microsoft, Apple, and CrowdStrike.

On June 9, 2026, Anthropic released Claude Fable 5: the first model from its "Mythos-class" tier — the level that now sits above the Opus class — that the company has cleared for general use. The framing from Anthropic this time is anything but modest. Anthropic's launch materials state that Fable 5 exceeds every model the company had previously made generally available, with its lead widening as tasks get longer and more complex.

In short: this is the most powerful AI model Anthropic has ever put in the hands of ordinary users. And it's available right now.

Claude Fable 5 vs. Claude Mythos 5: Two Models, One Engine

Here's where things get genuinely interesting — and a little nuanced.

Anthropic shipped two models simultaneously. Fable 5 is the public one, with safety classifiers turned on. Mythos 5 is the same underlying model with the guardrails lifted, locked behind an invitation-only program.

Think of it this way: Fable 5 and Mythos 5 are identical twins. Same brain, same training. The difference is that Fable 5 has a set of built-in safety gatekeepers that quietly intercept certain categories of sensitive requests — and route them to the weaker Claude Opus 4.8 instead.

Claude Mythos 5 will continue to be offered through Project Glasswing in partnership with the US government. Anthropic calls it the world's strongest cybersecurity model, scoring 78 percent on the ExploitBench benchmark — up from 69 percent for Mythos Preview and 40 percent for Opus 4.8.

Benchmark Breakdown: What the Numbers Actually Show

If you've spent the last year watching AI benchmarks inch forward by fractions of a point, Claude Fable 5 is a genuine jolt.

It posts 80.3% on SWE-Bench Pro while the next-best model sits 11 points behind. It finished a migration in a 50-million-line codebase in a single day. And AI researcher Andrej Karpathy called it "a major-version-bump-deserving step change forward."

Here's a quick snapshot of where Fable 5 leads the field:

Benchmark	Claude Fable 5	Claude Opus 4.8
SWE-Bench Pro (agentic coding)	80.3%	69.2%
Senior Engineer Benchmark	91/100	63/100
USAMO 2026 (mathematics)	97.6%*	42.3%
ExploitBench (cybersecurity)	40.0% (safeguarded)	40.0%
Terminal-Bench 2.1	88.0%*	—

*Scores marked with an asterisk reflect Mythos 5 performance; Fable 5 lands closer to Opus 4.8 on safeguarded topics.

Unlike models optimized mainly for short chat responses, Claude Fable 5 is designed for long-running, ambiguous, multi-step assignments — the kind of work that previously required a team.

When Anthropic gave the model persistent file-based memory while playing the deck-builder Slay the Spire, performance improved three times more than it did for Opus 4.8, and Fable reached the final act three times as often. That's Anthropic's way of illustrating something harder to benchmark: the model's ability to sustain focus and improve its own work over very long task horizons.

The Safeguards: How Anthropic Made a Dangerous Model Safe Enough to Ship

This is the part of the launch that has no real precedent in AI history. Anthropic didn't simply release a powerful model. It engineered a new kind of access framework — one built on automated classifiers rather than blanket restrictions.

The company introduced classifiers that detect certain categories of requests and route them to Claude Opus 4.8 instead of allowing Fable 5 to answer directly. The affected categories include cybersecurity, biology and chemistry, and attempts to extract model capabilities through distillation.

In practice, here's how each category works:

Cybersecurity: Anthropic ran more than 1,000 hours of external bug-bounty testing without producing a universal jailbreak, and one external partner rated Fable 5's protection against harmful cyber queries the strongest of any model tested.

Biology and chemistry: Anthropic says Fable 5 now routes most biology and chemistry requests to Opus 4.8, citing both stronger real-world scientific capability and concern about well-resourced bad actors.

Model distillation: The third category quietly catches requests aimed at "frontier LLM development" — essentially, using Fable 5 to help build a competing model. It's simultaneously a safety measure and a competitive one, and Anthropic is notably candid about baking both functions into the same mechanism.

The good news for most users: Anthropic says these fallbacks occur in fewer than 5% of sessions, meaning most users will effectively interact with the full Mythos-class model during ordinary use.

There is a friction point, though. Early testing by security researchers suggests the cyber safeguards may be broader than Anthropic's description implies. Rob T. Lee, chief AI officer and chief of research at SANS Institute, reported that his routine cybersecurity tasks involving incident response, detection, and basic forensic workflows were automatically routed from Fable 5 to Opus 4.8 during initial testing. Anthropic acknowledges the classifiers are intentionally broad and has pledged to narrow them over time.

Real-World Performance: What Early Users Are Saying

Benchmarks tell part of the story. Early production use tells the rest.

During testing, Stripe used Fable 5 to complete a codebase-wide migration in a single day — a task that would traditionally take a team of engineers weeks to plan and execute. That's the kind of real-world signal that turns benchmark numbers into business value.

Matthew Pines, testing frontier physics research, reported Fable 5 got nearly to where GPT-5.5 landed after four days — but in 36 hours. For researchers and knowledge workers, that's not just impressive; it's potentially transformational.

The model is also being praised for its handling of multimodal and tool-heavy tasks. Knowledge work, spatial reasoning, tool use, legal research, and health-related tasks all show the same pattern: a clear lead over Opus 4.8 and the broader competitive field.

Availability and Pricing

Who Can Access Claude Fable 5?

Fable 5 became available on June 9, 2026, through the Claude API and Amazon Bedrock, and is included free for Claude Pro, Max, Team, and Enterprise users during an introductory window from June 9 through June 22, 2026. After June 23, usage credits will be required until capacity fully expands.

It is available through the Amazon Bedrock APIs, in the Claude Platform on AWS, and delivers exceptional performance in software engineering, knowledge work tasks, and vision.

How Much Does It Cost?

Claude Fable 5 is priced at $10 per million input tokens and $50 per million output tokens — exactly double Claude Opus 4.8's rate of $5/$25. Batch pricing is $5/$25 per million tokens. Requests rerouted by safety classifiers bill at the lower Opus rates, and refused requests that produce no output are free.

Technical Specs at a Glance

Model ID: claude-fable-5
Context window: 1 million tokens
Max output: 128,000 tokens per request
Input pricing: $10 / million tokens
Output pricing: $50 / million tokens
Platforms: Claude.ai, Claude Code, Claude Desktop, Claude API, Amazon Bedrock

What This Means for the Future of AI Development

Claude Fable 5 isn't just a product launch. It's a statement about how frontier AI labs plan to navigate an increasingly uncomfortable reality: their most capable models are also their most dangerous ones.

Anthropic has packaged a frontier system for public use by gating it with deterministic classifiers and tiered trusted access rather than withholding it entirely — fallback instead of refusal, mandatory retention, and restricted tiers for the most sensitive cyber and biology work. The wider significance is that this approach now reads as a template other frontier labs facing the same dual-use dilemma may be pushed to follow.

The mandatory 30-day data retention policy for all Mythos-class model traffic is another signal. Anthropic is requiring 30-day data retention for all traffic on Mythos-class models, on both first- and third-party surfaces, to defend against multi-request attacks and to find false positives.

Whether this model for responsible frontier deployment actually works — whether the classifiers hold, whether jailbreaks emerge at scale, whether the biology safeguards prove too blunt — will define the next chapter of this story. But for now, Anthropic has done something no one else has: made the world's most capable AI model available to anyone willing to pay for a subscription, with a framework designed to keep the most dangerous capabilities out of the wrong hands.

Key Takeaways

Claude Fable 5 is Anthropic's first publicly available Mythos-class AI model, launched June 9, 2026.
It is the most capable model Anthropic has ever released to the general public, outperforming all previous models on coding, reasoning, and long-horizon tasks.
Fable 5 and Mythos 5 share the same underlying model; the difference is active safety classifiers in the public version.
Safety safeguards automatically route sensitive requests (cybersecurity, biology/chemistry, model distillation) to Claude Opus 4.8 — affecting fewer than 5% of sessions.
Pricing is $10/$50 per million tokens (double Opus 4.8), with free access for paid subscribers through June 22, 2026.
The launch sets a new industry precedent for how to deploy frontier AI models responsibly.

FAQs: Claude Fable 5 and the Mythos Release

Q: What is Claude Fable 5?
Claude Fable 5 is Anthropic's first publicly available Mythos-class AI model, released June 9, 2026. It is the most capable model Anthropic has ever made generally available, built for complex, long-horizon tasks like advanced coding, research, and multimodal knowledge work.

Q: How is Claude Fable 5 different from Claude Mythos 5?
They share the same underlying model. Fable 5 is the public version with safety classifiers active, which automatically route high-risk requests to Claude Opus 4.8. Mythos 5 has those cyber safeguards lifted and is restricted to vetted cybersecurity partners through Project Glasswing.

Q: Is Claude Fable 5 free to use?
Paid Claude subscribers (Pro, Max, Team, Enterprise) get free access to Fable 5 from June 9–22, 2026. After that, usage credits are required. API access is priced at $10 per million input tokens and $50 per million output tokens.

Q: Why do some of my requests get redirected to Opus 4.8?
Fable 5 has built-in safety classifiers that detect requests related to cybersecurity, biology and chemistry, and model distillation. When triggered, your request is automatically rerouted to Opus 4.8. Anthropic says this affects fewer than 5% of sessions and acknowledges some legitimate requests may be incorrectly caught.

Q: Can researchers and security professionals get full Mythos-class access?
Yes, in limited form. Cybersecurity organizations can apply through the Cyber Verification Program (CVP). Anthropic also plans a Trusted Access Program for biology researchers that will lift chemistry and biology safeguards while keeping cyber restrictions in place.

Google Snippet Suggestion

Claude Fable 5, released June 9, 2026, is Anthropic's first publicly available Mythos-class AI model — the most powerful Claude ever released to the public, featuring a 1M token context window, 80.3% on SWE-Bench Pro, and built-in safety classifiers that route sensitive requests to Opus 4.8.

Anthropic Releases Claude Fable 5: