Claude Fable 5: Anthropic's most capable model yet
Anthropic has released Claude Fable 5, the public version of the model it developed under the codename Mythos. Launched on June 9, 2026, Fable 5 becomes Anthropic's most capable generally available model and introduces a new "Mythos" tier that sits above the existing Opus line.
The headline gain is on hard, real-world engineering work. Fable 5 scores 80.3% on SWE-Bench Pro, a benchmark that measures a model's ability to resolve genuine software issues. For comparison, Anthropic's Claude Opus 4.8 scored 69.2% and OpenAI's GPT-5.5 scored 58.6% on the same test. Anthropic says the model's advantage tends to widen as tasks get longer and more complex.
Beyond coding, the company points to new highs in knowledge work, vision, memory and tool use. Those are the areas that matter most for the kind of multi-step, agentic work that AI tools are increasingly expected to handle on their own, from drafting and editing documents to analysing data and operating other software.
The public rollout ships with stronger safeguards designed to limit misuse while preserving the model's reasoning, coding and defensive-security strengths. Anthropic has also introduced a more powerful, restricted variant called Claude Mythos 5, available only to vetted partners through a program it calls Project Glasswing, alongside new pricing and broader research-access plans.
For anyone choosing between AI tools today, the takeaway is simple: the frontier moved again, and the gap between the leading models is now measured less by trivia and more by how reliably they can finish long, complicated jobs.