Anthropic Cuts Fable 5.1 Cache Costs by 75% as New Models Target Agentic Workloads

Quick Facts

• Cache reads for Fable 5.1 dropped to $0.25 per million tokens, a 75% cut from Fable 5’s $1.00 rate.
• Fable 5.1 scored 52.6% on Terminal-Bench-Science 0.1, beating Opus 5 at 29.0% and GPT-5.6 Sol at 22.4%.
• Enterprise partners including Jane Street, Cognition, Datadog, and MongoDB validated the model before launch.

Anthropic launched Claude Fable 5.1 and Claude Mythos 5.1 on Sept. 1, 2026, cutting cache-read prices sharply and posting benchmark scores that top its own prior models and OpenAI’s GPT-5.6 Sol.

The two models share the same underlying architecture. Fable 5.1 is generally available. Mythos 5.1 ships through Anthropic’s trusted access programs and carries safeguards built for cybersecurity and life sciences work.

What Fable 5.1 Costs

Standard pricing sits at $10 per million input tokens and $50 per million output tokens. Cache reads now run $0.25 per million tokens. Five-minute cache writes cost $12.50 per million and one-hour writes cost $20.

The $0.25 cache-read rate is 2.5% of the normal input price, well below the 10% multiplier Anthropic applies to most other Claude models. Anthropic says it cuts costs for typical workloads by about 25% and for complex coding and agentic tasks by up to 45%, based on four weeks of actual usage across Claude Enterprise, Claude Code, and the API during August 2026.

The math produces some counterintuitive comparisons. Fable 5.1’s cached input price is half of Opus 5’s cache-read rate, even though its base input price is twice as high.

Performance Numbers

On Terminal-Bench-Science 0.1, Fable 5.1 scored 52.6%, against 29.0% for Opus 5, 24.7% for Fable 5, and 22.4% for GPT-5.6 Sol. Anthropic puts the standard error at plus or minus 3.5 to 4.5 points per model.

On AutomationBench, the model scored 31.4%, up from 17.1% for Fable 5 and ahead of Opus 5’s 26.9%. On CursorBench 3.2.0, it reached 73.4%. GDPval-AA v2 knowledge work scores came in at 1,853 for Fable 5.1, compared with 1,824 for Opus 5 and 1,723 for Fable 5.

Both models support one million tokens of context and up to 128,000 output tokens per response.

What Enterprise Partners Report

Contract management firm results showed the model improved on Anthropic’s RedlineBench from 47.9 to 57.0 versus Fable 5, with the gain driven by first-turn quality, which doubled. Incident investigation evals at one partner showed Fable 5.1 outperforming Opus 5 at diagnosing complex production failures.

Cognition said it is moving Opus 5 traffic in its Devin product to Fable 5.1. The company measured task cost at $2.68 per task, down from $5.84 on Fable 5. Jane Street’s quantitative research team said the model solved more internal coding benchmarks than Fable 5 or Opus 5 while staying easier to follow across long multi-step tasks.

Artificial Analysis offered a different read. At maximum effort settings, it measured Fable 5.1 at $3.76 per completed task, about 20% above Fable 5, because the model produces roughly 1.7 times as many output tokens.

Enterprise Privacy Controls

Anthropic introduced what it calls Enterprise Frontier Safeguards. The system stores data in cloud infrastructure controlled by the customer, not Anthropic, and is designed to match the privacy protection of a zero data retention policy while blocking adversarial use.

Fable sits at the top of Anthropic’s model lineup. Below it are Opus, Sonnet, and Haiku as the large, medium, and small tiers. The 5.1 release builds on the Fable 5 and Mythos 5 models Anthropic shipped on June 9, 2026.

Read more: Anthropic’s Claude Fable 5.1 and Mythos 5.1 arrive with a 75% cost reduction for Fable cache reads