Photo: Ludovic Delot / Pexels
Anthropic's latest Claude model pushes ARC-AGI-2 coverage to 90% while slashing per-task costs and dramatically improving science benchmarks
Anthropic released Claude Fable 5.1 on September 1, positioning it as the company’s most capable and cost-efficient model for complex reasoning tasks. The headline number: 90% coverage on the ARC-AGI-2 benchmark at 32% less cost per task than its predecessor, Fable 5, which carried a verified price tag of $5.45 per task.
The benchmark gains in context #
Fable 5.1’s 90% coverage on ARC-AGI-2 represents a notable step up from Fable 5’s verified score of 89.2%. On Terminal-Bench 4.0, Fable 5.1 scored 55.8%, compared to Fable 5’s 42.0%. That’s a 13.8 percentage point jump, roughly a 33% improvement in raw scoring terms.
Terminal-Bench-Science 0.1, a recently introduced evaluation for scientific reasoning, told an even more striking story. Fable 5.1 posted 52.6% versus Fable 5’s 24.7%. The model more than doubled its predecessor’s performance on scientific tasks.
Fable 5.1 also topped multiple intelligence and agentic indexes for 2026, including an Artificial Analysis score of 66 under maximum effort conditions. The model ships with a 1 million token context window, 128K maximum output, and an adaptive reasoning system featuring five distinct effort levels.
Where the cost savings actually come from #
The 32% cost reduction per task is partly a story about smarter caching. Anthropic cut cache-read pricing by 75%, dropping it to $0.25 per million tokens. For typical operations, the cache pricing shift translates to roughly 25% savings. For complex, multi-step workflows where the model is reading from cache frequently, savings climb as high as 45%.
The base pricing structure itself remains unchanged. Standard tasks still run $10 per million tokens, and complex tasks cost $50 per million tokens.
Fable 5.1 arrives less than three months after Fable 5 launched on June 9, 2026. Fable 5.1 is available immediately through Anthropic’s API and major cloud providers. Meanwhile, Fable 5 is still undergoing testing on the ARC-AGI-3 benchmark, a reminder that verification processes for these advanced systems continue to evolve alongside the models themselves.
Disclosure: This article was edited by Editorial Team. For more information on how we create and review content, see our