June 10, 2026
On June 9, 2026, Anthropic released Claude Fable 5 and Claude Mythos 5, its first models in a new Mythos-class tier that sits above the Opus class. Claude Fable 5 is the version made generally available to the public, while Claude Mythos 5 is the same underlying model with some safeguards lifted, restricted to vetted partners. Anthropic reports that Claude Fable 5 reaches 95.0 percent on SWE-bench Verified and ships with a 1 million token context window, making it the company’s most capable model for coding and agentic work to date.
What is Claude Fable 5 and how does it differ from Mythos 5?
Claude Fable 5 is the first generally available Mythos-class model from Anthropic, a capability tier the company places above its existing Opus class. Anthropic says Claude Fable 5 and Claude Mythos 5 share the same underlying model, but the two are packaged differently for different audiences. Claude Fable 5 ships with production safeguards and is open to general API and product users under the model ID claude-fable-5.
Claude Mythos 5 is the unrestricted counterpart, with some safeguards removed in specific domains. Access to Mythos 5 is limited to vetted partners in Anthropic’s Project Glasswing program, which covers cyberdefenders and infrastructure providers. The split lets Anthropic put frontier capability in public hands through Fable 5 while keeping the higher-risk configuration of Mythos 5 behind a trusted-access gate.
Claude Fable 5 benchmarks and technical specs
Anthropic reports that Claude Fable 5 scores 95.0 percent on SWE-bench Verified and 80.3 percent on SWE-bench Pro, two widely used software engineering benchmarks. On coding-specific evaluations it reaches 29.3 percent on FrontierCode Diamond, 72.9 percent on CursorBench 3.1 at maximum effort, and 88.0 percent on Terminal-Bench 2.1.
The model supports a 1 million token context window by default and a maximum output of 128k tokens per request. It is exposed through the Claude API as claude-fable-5. These figures position Claude Fable 5 as a model aimed at long-context, multi-step coding and agentic tasks, where the combination of a large context window and high SWE-bench scores matters most.
How does Claude Fable 5 compare to Claude Opus 4.8, GPT-5.5, and Gemini 3.1 Pro?
On SWE-bench Pro, Anthropic reports Claude Fable 5 at 80.3 percent, ahead of its own Claude Opus 4.8 at 69.2 percent, OpenAI’s GPT-5.5 at 58.6 percent, and Google’s Gemini 3.1 Pro at 54.2 percent. The gap over Claude Opus 4.8, released on May 28, 2026, is the clearest signal that the Mythos class represents a real step up rather than a minor revision.
Claude Opus 4.8 remains in the lineup and continues to handle requests that Claude Fable 5 routes away for safety reasons. For organisations already standardised on Claude Opus 4.8, Claude Fable 5 is positioned as the higher-capability option for the hardest coding and agentic workloads, with Opus 4.8 serving as both a fallback and a lower-cost alternative.
Claude Fable 5 safety features and the Mythos-class tier
Claude Fable 5 ships with safety classifiers that route certain categories of request to Claude Opus 4.8 rather than answering them directly. According to Anthropic, requests touching cybersecurity, biology, chemistry, and model distillation fall back to Opus 4.8. This routing is the main practical difference between Fable 5 and the unrestricted Mythos 5.
The release comes days after Anthropic publicly warned that frontier AI is becoming more dangerous, and the company frames the Fable and Mythos split as a way to ship a Mythos-class model to the public while containing its highest-risk uses. Mythos 5 itself stays behind the Project Glasswing trusted-access program for defensive and infrastructure partners.
Claude Fable 5 availability and pricing
Claude Fable 5 is available now through the Claude API, Amazon Bedrock, Microsoft Foundry, and GitHub Copilot, and through partner products such as Harvey. Pricing is 10 US dollars per million input tokens and 50 US dollars per million output tokens, with batch pricing at 5 US dollars per million input tokens and 25 US dollars per million output tokens.
Claude Mythos 5 is not generally available and is limited to vetted Project Glasswing partners. For most teams, Claude Fable 5 is the model they will use, with Claude Opus 4.8 remaining available for workloads where its lower price or its handling of restricted categories is preferable.
Full details are in Anthropic’s Claude Fable 5 and Claude Mythos 5 announcement.
