Claude Fable 5: Anthropic Brings Mythos-Class AI Power to Everyone, Safely

Anthropic Just Made Its Most Powerful Model Available to Everyone

On June 9, 2026, Anthropic released Claude Fable 5, the most capable model the company has ever made generally available. Fable 5 is the first public member of the Mythos family — the frontier system that has already driven breakthroughs in defensive cybersecurity research and drug design earlier this year — and it arrives wrapped in a safety architecture designed from the start to make that power broadly accessible. Alongside it, Anthropic introduced Claude Mythos 5, the same underlying model with safeguards reserved for vetted partners. For anyone tracking the frontier of AI, this is one of the most consequential releases of the year — not only for what the model can do, but for how carefully it was brought to market.

What Claude Fable 5 Can Do

Fable 5 posts state-of-the-art results on nearly every capability benchmark Anthropic tested, and the software-engineering numbers are the headline. Stripe reported that the model compressed months of engineering into days, completing a 50-million-line codebase migration in a single day — work that would have taken a team roughly two months. Claude Fable 5 also earned the highest score on Cognition's FrontierCode evaluation and became the first model to break 90% on a demanding analytics benchmark, a ten-point jump over Claude Opus 4.8.

Vision and Long-Context Reasoning

The gains extend well beyond code. In vision tests, Fable 5 can reconstruct a web app's source from a single screenshot and read precise values straight off scientific figures. Its long-context reasoning holds focus across millions of tokens, and memory improvements tripled its performance on complex, multi-step tasks. On Hebbia's senior-level finance benchmark it took the top score, and Cursor's chief executive said the model "opened up a class of long-horizon problems that were out of reach."

Safety Built Into the Release, Not Bolted On

The most thoughtful part of this launch is architectural. Claude Fable 5 ships with three classifiers — covering cybersecurity, biology and chemistry, and model distillation. When a request touches one of those sensitive areas, Fable 5 routes the response to the more constrained Claude Opus 4.8 instead of answering directly. Anthropic tuned these safeguards conservatively, so they trigger in fewer than 5% of sessions on average, and more than 1,000 hours of red-teaming surfaced no universal jailbreaks. Mythos-class traffic is held for 30 days, never used for training, and then automatically deleted.

Why the Fallback Design Matters

This is a genuinely elegant way to release frontier capability responsibly. The vast majority of what people ask a model to do — writing code, analyzing data, synthesizing research, drafting documents — carries no special risk, and Fable 5 serves those requests at full strength. The classifier-and-fallback design means the safety considerations fall almost entirely on the narrow slice of genuinely high-risk queries, rather than gating the entire model behind enterprise contracts. Anthropic's automated alignment evaluation found misaligned behavior to be low and similar to that of Opus 4.8, and because Fable 5 shares Mythos 5's underlying model, it is expected to match. Readers who followed our AI Security coverage of Anthropic's Glasswing disclosure program will recognize the same defense-first philosophy at work here.

Pricing and Availability

Claude Fable 5 is priced at $10 per million input tokens and $50 per million output tokens, and it is available immediately through the Claude API, consumption-based Enterprise plans, Amazon Web Services, and GitHub Copilot. From June 9 through June 22, Anthropic is including Fable 5 at no extra cost for Pro, Max, Team, and seat-based Enterprise subscribers, with usage credits applying from June 23 onward. Claude Mythos 5 — the same model with safeguards lifted — is going to authorized partners through trusted-access programs for cybersecurity and biology researchers.

A Template for Releasing Frontier AI Responsibly

What makes Claude Fable 5 stand out is that it pairs a real capability jump with a release model that takes safety seriously without locking the technology away. The lesson for the wider field is an optimistic one: frontier capability and broad access do not have to be at odds when the safeguards are engineered in from day one. As more labs approach this level of capability, expect the fallback-classifier pattern Anthropic introduced with Fable 5 to become a widely studied reference point.

Sources: Anthropic, "Claude Fable 5 and Claude Mythos 5" (June 9, 2026); VentureBeat (June 9, 2026); Amazon Web Services Blog (June 9, 2026); GitHub Changelog (June 9, 2026).