Anthropic ships Fable 5.1 with a 75% cache read price cut
- Source
- x.com
- Date
Anthropic released Claude Fable 5.1 on September 1, but its sources entered our Intel corpus too late for that day's Brief. Artificial Analysis scored it at 66, the highest result on its Intelligence Index, while measuring $3.76 per task because it used about 1.7 times as many output tokens as Fable 5. Perplexity ranked it first on its August WANDR agent evaluation at 0.601 and $12.76 per task, a 21 percent score gain and 37 percent lower task cost than Fable 5. Cursor separately reported 73.4 percent at max effort on CursorBench 3.2. Adoption still has a constraint: GitHub says Fable 5.1 in Copilot enables data retention by default so Anthropic's safety classifiers can run, with only a temporary enterprise exception. Teams should compare task completion and data policy together before switching.
We’re introducing Claude Fable 5.1 and Claude Mythos 5.1. They're the world’s most advanced models for coding and knowledge work.
Fable 5.1 excels at complex, long-running tasks. And its research capabilities offer an early glimpse of how AI models will contribute to scientific progress.
Across our benchmarks, the model sets a new standard. It scores 52.6% on Terminal-Bench-Science 0.1, more than double Fable 5. On Terminal-Bench 4.0, it scores 55.8% against 42.0% for Fable 5.
As well as being capable of much higher performance than Fable 5, it can also achieve similar or better results at a much lower cost when set to lower effort levels.
Context
Anthropic's Fable 5.1 price change chiefly rewards workloads that repeatedly reuse already processed input. Its launch article prices cache reads at $0.25 per million , 75% below Fable 5, while ordinary input and output prices remain unchanged.
A 75% cut to one component produces a smaller reduction in the total bill. Anthropic estimates about 25% savings for typical workloads and up to roughly 45% for -heavy, tool-heavy work. Its comparison applies the prices to the same workloads from four weeks of August usage. The benefit therefore depends on how much of an application's processing consists of cached input.
- token — The chunk of text a model reads and writes in — roughly three-quarters of a word — and the unit AI usage is billed in.
- context window — The maximum amount of text a model can consider at once — its working memory for the current conversation or task.
Checking sign-in…
Loading comments…