Claude Fable 5.1 cuts agentic-workload cost by up to 45 percent, and Cognition is moving production traffic to it

Anthropic released Claude Fable 5.1 today, generally available on all platforms, alongside Claude Mythos 5.1, a restricted-access variant for vetted cybersecurity and life-sciences work. Cache reads, where the model reuses context it has already processed, now cost 75 percent less, at $0.25 per million tokens. Input and output pricing is unchanged from Fable 5, at $10 and $50 per million tokens. Because agentic work leans heavily on cache reads, Anthropic reports the overall cost of a typical workload drops about 25 percent, and highly agentic, context-heavy work drops by as much as 45 percent. On Anthropic's own benchmarks, Fable 5.1 also outscores Fable 5 and Opus 5 on agentic coding tasks: 55.8 percent versus 42.0 and 52.3 on Terminal-Bench 4.0, and 73.4 percent versus 70.5 and 70.0 on CursorBench 3.2.0. Walden Yan, Cognition's co-founder and chief product officer, said the company is moving its Devin coding agent off Opus 5 and onto Fable 5.1 starting on launch day: "With the new cache read pricing a Fable-class model is finally economical for the workloads we'd kept on Opus, starting with code review." Separately, Anthropic tightened its anti-distillation controls: API accounts created from today onward can no longer edit Claude's prior context in a conversation while keeping the transcript of its earlier thinking, closing a technique used to extract model behavior at scale.

Why it matters: Anthropic reports a cost reduction of up to 45 percent for cache-heavy agent workloads that make frequent tool calls and run for long periods. Cognition is already moving production traffic from a more expensive model. Anyone running agents on cost-sensitive workloads should re-run their model selection math this week.