Claude Fable/Mythos 5.1: Agent Economics Reshuffled

SOTA model + 75% cache cut − 70% more output = new cost math for builders.

The Numbers

Fable 5.1 leads every benchmark: 52.6% Terminal-Bench-Science, 55.8% Terminal-Bench 4.0. Cache reads: $0.25/MTok (75% off). Output: 1.7x more tokens → 20% net cost increase.

The Angle

Fable 5.1 is positioned for "complex, multi-step work that runs on its own." Same weights as Mythos, different routing. Grok 4.7 and Gemini Flash 3.8 incoming. World Labs Astra also dropped — world models are the other frontier.

For PromptEngines

Harness benchmarks need recalibration. Output verbosity is now a cost lever. One model, two personas — prompt engineering becomes the differentiator.