Anthropic released Claude Fable 5.1 on September 1, alongside a more restricted sibling, Claude Mythos 5.1, in what the company is positioning as an incremental but broad upgrade over the Fable 5 and Opus 5 models it shipped earlier this year. Fable 5.1 is generally available to all users; Mythos 5.1 is limited to U.S. companies and individuals in Anthropic's trusted access programs, reflecting its intended use in sensitive domains like cybersecurity and life sciences.

Benchmark gains and a pricing cut

Anthropic said Fable 5.1 now beats Opus 5 on every benchmark the company has published, including several where Opus 5 previously led — reversing the split performance profile the two models launched with in July. The clearest jump came on Terminal-Bench-Science 0.1, a benchmark for agentic scientific-research tasks, where Fable 5.1 scored 52.6% versus Fable 5's 24.7%. Anthropic also cited a roughly 13% improvement on the Terminal-Bench 4.0 coding benchmark and said Claude Code users should see about 60% fewer cybersecurity false positives.

Sticker pricing is unchanged at $10 per million input tokens and $50 per million output tokens, but cache-read pricing dropped 75%, from $1.00 to $0.25 per million tokens. Because heavily agentic workloads lean on cache reads for repeated context, Anthropic estimates that translates to roughly 25% lower costs for typical usage and up to 45% for agent-heavy applications. Fable 5.1 is also now able to help discover software vulnerabilities, though Anthropic says it is deliberately restricted from generating exploits for them.

Context: a $35 billion Lambda deal

The launch landed a day after Anthropic disclosed a $35 billion infrastructure agreement with cloud-compute startup Lambda, adding to a stack of large compute commitments the company has made through 2026 as it races to keep pace with rival labs' training and inference capacity. Fable 5.1 became available simultaneously through Anthropic's own API, AWS, and other major cloud partners.

Why it matters

The release keeps Anthropic on a roughly six-to-eight-week cadence of model updates through 2026, with the emphasis shifting from raw capability claims toward cost efficiency and coding/agentic reliability — the metrics enterprise buyers building on Claude Code and third-party agent frameworks care about most. A 75% cut to cache-read pricing is also a direct response to competitive pressure on inference costs from rivals like OpenAI, Google and Chinese labs such as DeepSeek and Moonshot AI, all of which have pushed aggressively on price this year.