Claude Fable 5.1

Anthropic announced Claude Fable 5.1 on 1 September 2026 alongside Claude Mythos 5.1, both the same underlying model separated only by guardrail levels. Fable 5.1 shipped in general availability under the API identifier claude-fable-5-1. It scores 52.6% on Terminal-Bench-Science 0.1 against 24.7% for Claude Fable 5 and 29.0% for Claude Opus 5, reaches 73.4% on CursorBench 3.2.0, and carries a one-million-token context window.

The pricing change landed on the cache: reads fell 75% to $0.25 per million tokens while input and output held at $10 and $50, which Anthropic puts at roughly 25% lower total cost on typical workloads and 45% on heavily agentic ones. Anthropic also authorised the model to discover software vulnerabilities while withholding exploit development, and introduced Enterprise Frontier Safeguards, which store customer data on the customer's own infrastructure.

Capability arrives with latency and access friction attached. Alan Blount measured Fable 5.1 producing a formal dependency graph for a Postgres to Spanner migration in 12 seconds where Gemini 3.8 Flash emitted a linear sequence in 3. On Google Cloud the model falls under the Advanced AI Safety Addendum, so a project that has not enabled prompt-response sharing through setPublisherModelConfig receives a 403 on every request.

Anthropic's threat report of 10 September 2026 lists preserved thinking, which prevents new API accounts from modifying the context preceding reasoning, among its countermeasures against distillation. Z.ai abandoned Fable for models it judged less protected.