Anthropic just rolled out Claude Fable 5, its first Mythos-class AI model for the public. It’s a big milestone for the company, especially as the race for advanced AI heats up.
The company revealed Fable 5 on June 9, 2026, calling it their most capable Claude model yet and making it widely available. They’re pitching it as a top choice for hard reasoning, long-term software engineering, deep research, knowledge work, and even vision tasks that mix different types of data.
Fable 5 officially sits in Anthropic’s new Mythos class, which outperforms the Opus tier. This release follows the earlier, invite-only Mythos Preview that Anthropic tested with cybersecurity partners through Project Glasswing.
So, what’s special about Fable 5? It doesn’t just move faster—it can work on tricky problems for longer stretches, acting more independently than any Claude before it. Anthropic says it shines in software engineering, data analysis, memory, knowledge work, and research, especially in the life sciences.
Still, Anthropic isn’t throwing caution to the wind. Fable 5 comes with some strong guardrails, especially on touchy subjects like cybersecurity, chemistry, and biology. Sometimes those protections might block a harmless request, but Anthropic expects that to happen less than 5% of the time.
If a request seems too sensitive, Fable 5 won’t handle it directly. Instead, it either declines or hands you off to a more restricted model like Claude Opus 4.8, to keep things safe while still letting people use cutting-edge AI.
Alongside Fable 5, Anthropic also introduced Claude Mythos 5—built on the same foundation, but with some safeguards removed. Mythos 5 is strictly invite-only for now, aimed at selected Project Glasswing partners like cybersecurity and infrastructure groups as a step up from the previous Mythos Preview.
Project Glasswing itself is Anthropic’s program for helping defenders secure vital software infrastructure using advanced AI. The goal is simple: Give trusted experts access to stronger security tools before they go mainstream.
According to Anthropic, Mythos 5 sets the new bar for AI cybersecurity and they’ll expand access gradually through a trusted-accounts system. Plus, there’s talk of allowing select biology researchers to use Fable 5 with more relaxed biology and chemistry restrictions in the future—with cybersecurity protections staying in place.
Pricing is straightforward: $10 per million input tokens, $50 per million output tokens for both models. That’s less than half what users paid for Mythos Preview.
Developers can get Claude Fable 5 via the Claude API and platforms like AWS, Amazon Bedrock, Vertex AI, and Microsoft Foundry. Mythos 5, though, stays exclusive to approved Glasswing users.
There’s also a launch perk: From now until June 22, Fable 5 comes standard in Anthropic’s Pro, Max, Team, and Enterprise seat plans at no extra cost. After June 22, you’ll need usage credits unless Anthropic decides to extend the window. The company says they’ll bring Fable 5 back as a permanent feature for subscribers as soon as they can handle more demand.
This launch happens as AI companies struggle to strike a balance between rolling out powerful new tools and keeping safety in check. Anthropic is drawing a clear line: the broader public gets a strong, safeguarded model in Fable 5, while ultra-capable, less-restricted tech is reserved for trusted defenders and infrastructure players.
For a lot of people—businesses, developers, scientists—Claude Fable 5 is a big leap forward. But for policymakers and security pros, the bigger question is how companies can launch ever more powerful AI without risking it falling into the wrong hands.
For now, Anthropic’s answer is to keep tight controls, enforce careful safety features, and run separate access tracks for public users and trusted domain experts.
