Claude Fable 5: Anthropic Releases Its Most Powerful Public Model

On June 9, 2026, Anthropic launched Claude Fable 5, the most capable model it has ever made generally available. Alongside it came a locked-down sibling, Claude Mythos 5, reserved for cyberdefenders. Here’s what the new model does, how Anthropic kept it safe, and what it costs.

Claude Fable 5 is what Anthropic calls a “Mythos-class” model that has been made safe for general use. In plain terms, it is the most powerful model the company has ever released to the public.

According to Anthropic, Fable 5 reaches state-of-the-art results on nearly every benchmark it tested, spanning software engineering, knowledge work, vision and scientific research. Notably, the company says the longer and more complex the task, the larger Fable 5’s lead grows over its other models. It can also work autonomously for longer than any previous Claude.

The two models share the same brain. Claude Fable 5 is the public, safeguarded version, while Claude Mythos 5 is the same underlying model with some safeguards lifted.

Because of that raw power, Mythos 5 stays restricted. Anthropic is offering it only to a small group of cyberdefenders and infrastructure providers through Project Glasswing, a collaboration with the US government. The program is expanding to roughly 150 organizations across more than 15 countries, with select biology researchers next in line. Anthropic claims Mythos 5 has the strongest cybersecurity capabilities of any model in the world.

The numbers behind Claude Fable 5 are hard to ignore. It posts the highest score on Hebbia’s finance benchmark, and the trading firm IMC said it aced their trading-analysis evaluations. On vision, it even played Pokémon FireRed using sight alone.

Anthropic also points to gains in knowledge work and scientific research, areas where earlier models often stumbled on multi-step reasoning. Meanwhile, the restricted Mythos 5 reportedly accelerated protein design by around ten times. Taken together, the results suggest Anthropic has pushed well past its previous frontier.

Benchmarks are one thing; real work is another. During early testing, Stripe reported that Claude Fable 5 compressed months of engineering into days. In one case, it completed a 50-million-line codebase migration in a single day — a job that would otherwise have taken a full team more than two months.

As a result, developers are among the first to feel the difference, especially on long, multi-step tasks where the model can keep going on its own.

To release such a capable model publicly, Anthropic wrapped Claude Fable 5 in three safety classifier systems. They cover cybersecurity, biology and chemistry, and model distillation, blocking high-risk requests in each area.

When a safeguard triggers, the query is quietly routed to Claude Opus 4.8 instead. Importantly, Anthropic says this happens in fewer than 5% of sessions on average. The distillation classifier, meanwhile, is meant to stop anyone from extracting Fable 5’s capabilities to train a rival model. External red-teamers also threw 30 jailbreak techniques at the cyber safeguards, and according to the company, zero harmful single-turn requests got through.

Pricing is surprisingly aggressive. Claude Fable 5 costs $10 per million input tokens and $50 per million output tokens.

That is less than half the price of the earlier Claude Mythos Preview. In other words, the most powerful public Claude is also one of the better values for heavy, token-hungry workloads.

Access is broad from day one. Claude Fable 5 is available immediately on the Claude API and consumption-based enterprise plans, and it has already rolled out to GitHub Copilot and Amazon Bedrock.

For subscribers, Anthropic is including Fable 5 at no extra cost from June 9 to 22. After June 23, it will draw on usage credits, with a return to standard plans promised as quickly as the company can manage.

The timing is pointed. Anthropic shipped Claude Fable 5 just days after warning that AI is getting too dangerous, and two months after the private Mythos rollout rattled Wall Street.

In short, the company is threading a needle: ship its strongest public model while keeping the most sensitive capabilities locked behind Project Glasswing. Rivals such as OpenAI and Google are racing on the same frontier, so the pressure to release is only growing. Whether that balance between capability and caution holds will help shape the next phase of the AI race.

If a smarter Claude has you rethinking your stack, explore our picks for the best AI coding tools and the best AI writing tools to put models like Claude Fable 5 to work.

*Sources: Anthropic, The Verge, TechCrunch, CNBC, Wired.*