Claude Fable 5: Mythos Power for General Use
Fable 5 is the first Mythos-class model cleared for normal users, with big leads on long-horizon coding and research at double the prior price and with new safeguard fallbacks.
Anthropic shipped Claude Fable 5 on June 9 as the first Mythos-class model available to ordinary users. It shares the same base as the restricted Mythos 5 but adds conservative safety classifiers that hand off certain queries to Opus 4.8. The result is a model whose headline numbers sit well above prior flagships on the work that matters for engineering and research.
On Anthropic's coding evaluations Fable 5 reaches 95.0 percent on SWE-bench Verified and 80 percent on SWE-bench Pro, against 88.6 and 69.2 percent for Opus 4.8. It also leads the GDPval-AA knowledge-work Elo at 1932 versus 1890. Early testers report the margin widens on long, multi-step jobs; the harder the task, the more the gap shows up in practice.
Pricing doubles the previous tier: ten dollars per million input tokens and fifty for output. Context stays at one million tokens with no long-context surcharge. The model is slower and burns more tokens during heavy reasoning, so real workloads cost several times an Opus 4.8 session even before the rate card. Teams that only need occasional frontier performance will route most traffic elsewhere.
The safeguards are deliberately broad for launch. Cybersecurity, high-risk biology and chemistry, and distillation attempts trigger fallback; the average session sees no swap, but some innocent prompts still route to the older model and users receive notice. Anthropic states this keeps the dangerous edges contained while the rest of the model runs at full strength.
Fable 5 closes the loop that started with the April Mythos Preview decision to withhold the model. The same underlying engine is now priced for paid users with guardrails that are tunable over time. What changes for builders is access to a higher capability band for the hardest slices of agentic work, at a cost that forces explicit choices about when the extra performance is worth the spend.