Anthropic Opens Access to Its Most Powerful AI Model Yet-But Not for Long

After months of speculation and careful deliberation, Anthropic has finally released the first publicly usable model from its new Mythos family: Claude Fable 5. The company had been testing the more powerful Claude Mythos Preview since April, but concerns about its ability to detect-and potentially exploit-cybersecurity vulnerabilities kept it under wraps. Now, Fable 5 arrives as a safer, more controlled alternative.

According to a detailed blog post, Fable 5 is Anthropic's most advanced AI model to date while still maintaining strong safety guardrails. It offers notable improvements in software engineering, knowledge work, and image understanding. In a quirky benchmark, the model even managed to beat Pokemon FireRed using only a vision-based AI interface-a feat previous models couldn't pull off.

But there's a catch: Fable 5 requires double the computing power of Anthropic's Opus models, so the company is rolling it out cautiously. Starting Tuesday, subscribers on Pro, Max, Team, and Enterprise plans can use Fable 5 until June 22. After that, the model disappears from the standard lineup, though users can still access it by burning usage credits-a pay-as-you-go system that may add extra fees if you exceed your allotment.

Safety was a major factor in the rollout. Anthropic says Fable 5 rejected every harmful single-turn request related to cyberattack planning, exploit development, and defense evasion, even when tested with 30 different jailbreak techniques. To prevent bioterror risks, Fable 5 won't answer most biology or chemistry questions; those are routed to Opus 4.8 instead. The company is also working with the U.S. government on future plans for the full Mythos 5 release.

For developers, Fable 5 is available now through the Claude API at $10 per million input tokens and $50 per million output tokens. Anthropic hasn't set a timeline for when-or if-Fable 5 will become a permanent part of its subscription plans.