Anthropic has released Claude Fable 5.1.
The internals are the same model as Mythos 5.1, the only difference being the strength of the safety device. It is about 25% cheaper and more than twice as expensive as GPT-5.6 Sol for scientific research tasks. Although it may seem to be a matter of performance, the essence is the design of ``what is allowed to whom.'' Reactions overseas are divided in two depending on the severity of the safety measures.
Claude Fable 5.1 shipped Sept 1.
Terminal-Bench-Science: 52.6% vs Fable 5's 24.7% — more than double. Beats Opus 5 (29.0%) on it too. Base pricing unchanged at $10/$50 per million tokens, but cache reads fell $1.00 -> $0.25, ~25% typical savings and up to 45% agentic.
Fable 5 has always pleased me.
Opus 5, on the other hand, is a disappointment. Fable 5 helped me create this game. Appstore and Google Play on now
Fable 5.1 one-shotted this pretty video w/ a few hints at what’s coming next at Neurobro 👀
Been testing it a bit & my first impression is pretty clear: PROS: • Much better at following instructions. You ask for something specific - Fable gets it! • MUCH faster. Iterations
You have reached the end of the archive
All of Claude Opus 5