Fable 5 Benchmarks: What's Confirmed and What Isn't

The honest state of Fable 5 benchmarks

If you are looking for a tidy table of Fable 5 benchmark scores, the responsible answer is: there isn't a verified one yet. Anthropic has not published an official system card with benchmark results for the Mythos-class models. That means any specific number you see, a coding score, a reasoning percentage, an agentic-task pass rate, is unofficial until Anthropic confirms it.

Why no official numbers yet

Mythos-class launched on June 9, 2026, and a US export-control directive suspended access just days later, on June 12. The combination of a very new release and an active suspension means the usual cadence of system-card publication and independent evaluation has been disrupted. Treat the benchmark picture as still forming.

How to read unofficial scores

• Check the source. Community roundups and leaderboards are useful signals but are not vendor-confirmed.
• Watch the methodology. Prompting style, temperature, tool access, and harness all move scores. Two "Fable 5" numbers can disagree simply because the test setups differ.
• Remember Fable 5 and Mythos 5 share a model. Capability differences between them come from safeguards (classifier rerouting), not from a different base model, so benchmark behavior can diverge when a classifier reroutes a flagged request to Opus 4.8.

Where official numbers will appear

When Anthropic publishes the system card, that is the authoritative source for benchmark results. Until then, label every figure as unofficial and avoid citing specific percentages as fact.