Anthropic Just Released Its Most Powerful AI Model to the Public

Anthropic today launched two new AI models, Claude Fable 5 and Claude Mythos 5, marking the company's first broad release of the powerful "Mythos-class" AI capabilities it previously made available only to participating organizations in its restricted cybersecurity program, Project Glasswing.

Fable 5's capabilities exceed those of any model Anthropic has ever made generally available. It is state-of-the-art on nearly all tested benchmarks of AI capability, showing exceptional performance in software engineering, knowledge work, vision, scientific research, and many other areas.

The timing is notable. Claude Fable 5 arrives just days after Anthropic confidentially filed its IPO prospectus with the SEC, as the company's revenue run rate has ballooned to $47 billion, up from roughly $10 billion in annual revenue last year. This is a company building momentum fast, and Fable 5 is the most visible proof point.

What Is the Mythos Class?

Mythos-class models are a tier of Claude models that sit above the Opus class in capability. The first, Claude Mythos Preview, was released in April through Project Glasswing. That is followed today by Claude Fable 5 and Claude Mythos 5.

Anthropic captivated Wall Street and government officials in April with the unveiling of Mythos, which excels at identifying security flaws within software. The company did not plan to make the model generally available, limiting the rollout to a select group of companies as part of Project Glasswing.

Anthropic created Glasswing after internal evaluations showed Mythos-class models could find and exploit software vulnerabilities at a level that raised meaningful misuse concerns. The question was never whether to release it more broadly, but how.

Fable 5 vs. Mythos 5: The Key Difference

These two models share the same underlying architecture. They are the same underlying model. The distinction is access control. Claude Mythos 5 has no safety classifiers on cybersecurity and biology queries. Claude Fable 5 has safety classifiers layered on top.

Fable is from the Latin fabula, "that which is told," akin to the Greek mythos. The safeguards are what distinguish the two models and are why Anthropic gave them different names.

In practice, this means:

• The general-purpose Fable 5 wraps the same underlying Mythos-class capability in new safeguards. Requests involving certain high-risk areas, including cybersecurity, biology and chemistry, and model distillation, are automatically routed to Claude Opus 4.8 instead, with users notified when that happens.
• Mythos 5 is the same underlying model as Fable 5 but with the safeguards lifted in some areas. It has the strongest cybersecurity capabilities of any model in the world.
• To release the model both safely and quickly, Anthropic tuned these safeguards conservatively. They'll sometimes catch harmless requests, though they trigger, on average, in less than 5% of sessions.

Performance and Benchmarks

The benchmark numbers are hard to ignore. On SWE-bench Pro, which measures a model's ability to complete difficult software engineering tasks, Anthropic says Fable 5 and Mythos 5 reach 80.3%, vastly outperforming OpenAI's latest general model GPT-5.5, which scored 58.6%.

On Cognition's FrontierCode Diamond benchmark, which tests high-quality, maintainable agentic coding, the models score 29.3%, compared with 13.4% for Claude Opus 4.8 and 5.7% for GPT-5.5.

Third-party evaluators are seeing similar results. Analytics company Hex said "Fable 5 is the first to break 90% on our core analytics benchmark of complex, long-running analytical tasks — a 10-point jump over Opus." Legal AI platform Harvey also confirmed Fable 5 scored new highs on both LAB and BigLaw Bench, their benchmarks for evaluating model performance on complex legal tasks.

On efficiency, Claude Fable 5 beats Opus 4.8 on Anthropic's everyday spreadsheet suite at every effort level, finishing runs 25 to 30% faster. For long-running agentic workflows, one early customer found Fable 5 completed a frontier physics research task in 36 hours using one-third the reasoning tokens it took GPT-5.5 four days to match.

On the science side, Mythos 5 matched or exceeded skilled human operators on protein design tasks across 9 of 14 targets, completing work that normally requires a scientist to choose binding sites, run design tools, and handle failures autonomously.

Pricing and Access

Both models carry the API identifier claude-fable-5 and claude-mythos-5 respectively, with a 1M token context window, up to 128k output tokens per request, and pricing of $10 per million input tokens and $50 per million output tokens. Batch pricing drops to $5/$25 per million tokens.

That price ranks Fable 5 as the most expensive of major AI models available globally. It is, however, less than half the price of Claude Mythos Preview, which ran at approximately $30/$150 per million tokens.

For subscription users, the rollout has a temporary structure. Fable 5 will be included on Pro, Max, Team, and seat-based Enterprise plans at no extra cost through June 22. On June 23, the company plans to remove Fable 5 from those plans, after which using it will require usage credits. Anthropic aims to restore Fable 5 as a standard part of subscription plans as quickly as possible.

Claude Fable 5 is also now generally available in GitHub Copilot, the first model in Anthropic's Mythos class designed for long-horizon, autonomous coding and knowledge-work tasks. It's also live on Amazon Bedrock.

Safety Architecture and Data Retention

Anthropic stress-tested the classifiers before shipping. Wary of what a Mythos-class model could do in the wrong hands, Anthropic ran an external bug bounty that produced no universal jailbreaks in over 1,000 hours of testing, then worked with external red-teaming organizations which also failed to find universal jailbreaks.

One significant policy change accompanies the launch. With Fable 5 and Mythos 5, Anthropic will require a 30-day retention on all traffic, even if enterprises previously had zero-retention agreements. The company says it won't use the data for training and will use it only to defend against complex and novel attacks and identify false positives.

If your company has strict data-handling rules, this may decide whether Fable 5 can be used at all. On AWS, Bedrock users must opt into provider data sharing before invoking Fable 5, and regulated teams should route sensitive prompts carefully and keep Opus 4.8 available where retention is not acceptable.

Final Thoughts

What stands out technically is the fallback architecture. Rather than a hard block, Anthropic built a tiered response system where high-risk queries silently route to Opus 4.8 at Opus 4.8 pricing. It's a pragmatic engineering decision, and it lets Anthropic ship a genuinely capable model without waiting for perfect classifiers. The tradeoff is that some legitimate biology and chemistry queries will get downgraded, which will frustrate researchers until Anthropic narrows those classifiers.

The SWE-bench Pro numbers deserve attention. An 80.3% score versus GPT-5.5's 58.6% is not a marginal win on a coding benchmark, and the FrontierCode Diamond gap is even wider. If those numbers hold up under independent evaluation, Fable 5 makes a credible case for being the strongest coding model available to the public today. The 30-day mandatory data retention is the thing I'd watch most closely for enterprise adoption. Many teams that previously relied on zero-retention agreements will need to re-evaluate their workflows before deploying this at scale.

As always, run your own evals during the free window before June 22. What do you think about Anthropic's approach to releasing Mythos-class AI with safeguards? Drop your thoughts in the comments.

---