Anthropic Just Dropped Claude Fable 5 and Mythos 5 – Here Is What Matters

Anthropic Just Dropped Claude Fable 5 and Mythos 5 - Here Is What Matters

Anthropic launched Claude Fable 5 today, a Mythos-class model they have made safe for general use. It is state-of-the-art on nearly every benchmark they tested, and the pricing is surprisingly aggressive: $10 per million input tokens and $50 per million output tokens. That is less than half what Claude Mythos Preview cost.

The catch? Fable 5 comes with new safety classifiers that redirect some queries to Claude Opus 4.8 instead. If you ask about cybersecurity, biology, chemistry, or distillation, you get Opus 4.8. Anthropic says this happens in less than 5% of sessions, but they tuned it conservatively so benign requests will occasionally get caught. They plan to reduce false positives quickly.

For a small group of cyberdefenders and infrastructure providers, Anthropic also launched Claude Mythos 5. Same underlying model, fewer safeguards. It is the strongest cybersecurity model in the world and deploys through Project Glasswing with the US Government.

What Fable 5 Actually Does Better

The longer and more complex the task, the bigger Fable 5's lead. Here is where it stands out:

Software engineering. Stripe reported that Fable 5 compressed months of engineering into days. In a 50-million-line Ruby codebase, it performed a codebase-wide migration in a day that would have taken a team over two months manually. On Cognition's FrontierCode evaluation, which tests high-quality agentic coding, Fable 5 scores highest among frontier models even at medium effort.

Knowledge work. On Hebbia's Finance Benchmark for senior-level reasoning, Fable 5 has the highest score of any model. IMC noted it aced their trading-analysis evaluations across factual lookup, conceptual reasoning, root-cause analysis, and expected-value analysis.

Vision. Fable 5 is the new state-of-the-art for vision tasks. It can extract precise numbers from scientific figures and rebuild a web app's source code from screenshots alone. It also beat Pokemon FireRed with a minimal vision-only harness, something earlier Claude models could not do even with complex helper tools.

Memory and long context. Fable 5 stays focused across millions of tokens. In Slay the Spire, persistent file-based memory improved its performance three times more than it did for Opus 4.8, and it reached the final act three times more often.

Life sciences. Using Mythos 5, internal protein design experts accelerated drug design by around ten times. The model executed all tasks normally done by a scientist: choosing binding sites, running protein design tools, and recovering from failures. Nine of fourteen protein targets yielded strong candidates currently under investigation.

Novel research. Mythos 5 is Anthropic's first model to consistently produce novel, compelling scientific hypotheses. In blinded comparisons, scientists preferred Mythos's molecular biology hypotheses about 80% of the time. One hypothesis about an E. coli protein was independently corroborated by another lab.

The Benchmark Picture

Pricing and Access

Fable 5 and Mythos 5 are priced at $10 per million input tokens and $50 per million output tokens. That is a significant drop from Mythos Preview pricing. Anthropic is clearly trying to make advanced capabilities accessible quickly while keeping safety guardrails in place.

For most users, Fable 5 is the model you will actually use. Mythos 5 is restricted to trusted cyberdefenders and infrastructure providers through Project Glasswing, with broader access planned later.

The Safety Tradeoff

The new classifiers are the big story here. Anthropic is admitting that a model this capable poses real risks in cybersecurity and biology. Rather than refusing outright, they redirect sensitive queries to Opus 4.8. This is a pragmatic approach, but the conservative tuning means some harmless requests will get caught. Anthropic says they are working to reduce false positives fast.

The alignment assessment found that Mythos 5's level of misaligned behavior, including deception and cooperation with misuse, was low and similar to Opus 4.8. Since Fable 5 is the same underlying model, its alignment profile should be similar.

What I Think

This is Anthropic's most capable public release yet. The fact that they are shipping it with safety redirects rather than keeping it gated entirely suggests they believe the safeguards are strong enough for general use. The pricing undercuts Mythos Preview by more than half, which is a clear signal they want broad adoption.

The vision-only Pokemon run and the autonomous Factorio play are the demos that stick with me. Those are not benchmark numbers, they are proof that the model can sustain long-horizon tasks with minimal scaffolding. That is what matters for real agents.

For developers, the FrontierCode and codebase migration results are the headline. If Fable 5 can genuinely compress months into days on real codebases, that changes the economics of software engineering.

The safety redirect mechanism is a reasonable compromise for a model this capable, but the false positive rate will be the thing to watch. If Anthropic can get it below 1% without opening misuse vectors, they will have threaded the needle.

Source: https://www.anthropic.com/news/claude-fable-5-mythos-5