Anthropic Opens Access to Its Most Advanced AI Model - With Strings Attached

Anthropic is putting its most capable AI model into wider circulation, but the company is taking no chances on safety. On Tuesday, the AI firm launched Claude Fable 5, the first publicly available version of its Mythos model. Fable 5 is designed to handle demanding tasks in software engineering, knowledge work, and computer vision, but it comes with hard safety limits baked in. When it detects high-risk queries involving cybersecurity, biology, chemistry, or model distillation, the system blocks the response and hands off to the more conservative Claude Opus 4.8.

Mythos was first released as a preview in April, restricted to a handful of partners over cybersecurity concerns. Last week, Anthropic expanded access to hundreds of critical infrastructure organizations across 15 countries. Now, anyone can use Fable 5 through Anthropic's API or consumption-based Enterprise plans. Subscription users get it free through June 22 on Pro, Max, Team, and seat-based Enterprise plans. After that, Anthropic will move Fable 5 to a usage-credit model, with plans to restore it as a standard subscription feature as soon as possible.

The company is also rolling out Mythos 5 to organizations already approved for the advanced model. The launch comes as Anthropic prepares to go public alongside OpenAI and SpaceX, and follows its public call for major AI labs to establish a coordinated emergency brake on frontier AI development. Anthropic warns that systems are advancing so quickly they may soon achieve recursive self-improvement - autonomously upgrading themselves without human input.

To prevent misuse, Anthropic stress-tested Fable 5 with jailbreak attempts. It ran an external bug bounty that produced no universal jailbreaks in over 1,000 hours of testing, then worked with external red-teaming organizations that also came up empty. Still, novel attacks remain possible. As a result, Anthropic now requires a 30-day retention on all traffic for both Fable 5 and Mythos 5, even for enterprises that previously had zero-retention agreements. The data won't be used for training, only to defend against new jailbreaks and reduce false positives. The policy could set a precedent where access to powerful models comes with mandatory data retention framed as a safety measure.

In practice, most queries will get a Fable 5 response. Early data shows at least 95% of sessions run entirely on the model's own answers, with only rare deferrals to Opus 4.8. In third-party testing, analytics company Hex reported Fable was the first model to score 90% on its core benchmark of complex, long-running analytical tasks. "On the hardest questions, it shows strong judgment and attention to nuance," Hex said. Vibe-coding platform Base44 noted Fable excels at "one-shotting full apps" and has excellent tool-calling. AI workspace provider Genspark said Fable beat every other model in its evaluations, particularly on UI design and game coding.

Pricing is $10 per million input tokens and $50 per million output tokens - double the cost of Opus 4.8. That alone may limit widespread adoption. Many enterprises are already feeling the sting of AI costs, blowing through annual budgets early. Advanced models like Opus 4.8 can compound the problem by splitting a single request into multiple reasoning tasks. But some companies see the value. Shopping rewards platform Rakuten said, "At the highest effort, Fable reflects on and validates its own work. For us, that's what makes highly autonomous operations possible - the extra thinking pays for itself."

Source: TechCrunch