Anthropic Launches Claude Fable 5: Its Most Capable Model Yet

Anthropic released Claude Fable 5 today, calling it the most capable model it has ever made generally available. Fable 5 is the first model from Anthropic's new "Mythos" tier — a class that sits above its Opus models — to be opened up for everyday use.

The company says Fable 5 is state-of-the-art on nearly every benchmark it tested, with the biggest leads showing up on long, complex tasks. Alongside it, Anthropic launched Claude Mythos 5, the same underlying model with some safety limits removed, restricted to a small group of vetted cybersecurity partners.

What came before

This is the next step in a rollout Anthropic started in April. Back then, it introduced its first Mythos-class model — Claude Mythos Preview — but only to a limited circle of cyber defenders and critical infrastructure providers through a program called Project Glasswing.

The reason for the caution: Mythos-class models are powerful enough to be dangerous in the wrong hands. Anthropic said at the time that it hoped to eventually release Mythos-level capabilities to everyone, but only once it had built safeguards strong enough to reliably prevent misuse. Today's launch is the company saying those safeguards are finally ready.

What Fable 5 can actually do

The early results are striking. During testing, payments company Stripe reported that Fable 5 compressed months of engineering work into days — running a codebase-wide migration across a 50-million-line codebase in a single day, a job that would normally take a team over two months by hand.

Why this matters: code migrations are exactly the kind of slow, expensive, error-prone work that eats engineering teams alive. A model that can do it reliably in a day changes the math for every software company.

Fable 5 also posted gains outside of coding. On finance-reasoning tests from firms like Hebbia and IMC, it topped every other model. It set a new high mark on vision tasks — it can rebuild a web app's source code from screenshots alone, and it beat the video game Pokémon FireRed using only raw game images, with no maps or helper tools (something earlier Claude models couldn't manage).

The model can also stay focused across millions of tokens of context — meaning it can keep its place on tasks that run for hours without losing the thread. Customers including Cursor, GitHub, and Cognition described it as a clear step up for long-horizon, autonomous work.

The safety catch

Here's the trade-off Anthropic built in. Because Fable 5 is so capable, it could in theory help bad actors with things like cyberattacks or dangerous biology research. So the company wrapped it in a set of "classifiers" — separate AI systems that watch for risky requests.

When Fable 5 detects a query touching cybersecurity, biology and chemistry, or model "distillation" (copying its abilities to train rival models), it automatically hands the response to Claude Opus 4.8, Anthropic's next-most-capable model, instead. Users get told when this happens.

Anthropic admits the safeguards are tuned conservatively, so they'll sometimes catch harmless questions. But it says more than 95% of sessions involve no fallback at all — and for those, Fable 5 performs essentially the same as the unrestricted Mythos 5. The company ran an external bug bounty that logged over 1,000 hours of testing without producing a universal jailbreak.

Why the industry should care

This launch sharpens an already fierce model race. Anthropic is positioning Fable 5 against frontier models from OpenAI and Google, and the early customer quotes lean hard on one theme: autonomy. Companies are increasingly handing entire workflows — not just snippets — to AI agents, and Fable 5 is being pitched as the model that makes that trustworthy.

It also doubles as a statement about Anthropic's safety-first brand. By shipping cutting-edge capability and conservative guardrails in the same release, the company is arguing you don't have to choose between the two. Whether enterprise customers find the false-positive rate acceptable will be the real test.

There's a research angle too. Using Mythos 5, Anthropic's own scientists said they accelerated parts of drug design roughly tenfold, and the model generated novel molecular-biology hypotheses that its researchers preferred over rival models about 80% of the time in blind comparisons.

How to access it — and the fine print on pricing

Claude Fable 5 is available everywhere starting today through the Claude API at $10 per million input tokens and $50 per million output tokens.

For subscribers, the rollout is staged. From today through June 22, Fable 5 is included on Pro, Max, Team, and seat-based Enterprise plans at no extra cost. On June 23, Anthropic will pull it from those plans, and using it after that will require usage credits — a move driven by uncertain demand and capacity limits. The company says it aims to restore Fable 5 as a standard plan feature "as quickly as we can."

The bottom line: Anthropic just put its most powerful model in the hands of regular users, and the early benchmarks suggest a real jump in what AI can do on long, complex work. If you're a developer or a business leaning on AI agents, it's worth testing now — just keep an eye on that June 23 pricing change before you build it into your workflow.