Claude Fable 5.1 Goes Public with 75% Cheaper Cache Reads and Major Benchmark Gains

Anthropic released Claude Fable 5.1 as a generally available model today, cutting cache read costs by 75% and posting a 52.6% score on the Terminal-Bench-Science 0.1 benchmark. The same underlying model also powers Claude Mythos 5.1, which stays restricted to vetted US organizations under Project Glasswing. For AI builders, the headline changes are a much cheaper cache path for long-context agentic workloads and a clear separation between public and guarded model tiers.

Fable 5.1 goes public with a 75% cache read price cut

Claude Fable 5.1 is available as claude-fable-5-1 on the Claude API, Amazon Bedrock, Claude Platform on AWS, Google Cloud, and Microsoft Foundry. Base input and output pricing remains unchanged at $10 and $50 per million tokens, with batch processing at $5 and $25 per million tokens. The big change is cache reads: they drop from $1.00 to $0.25 per million tokens, a 75% reduction. Anthropic estimates this translates to roughly 25% lower cost on typical workloads and up to 45% on context-heavy agentic tasks. Both models support a 1M token context window and 128K max output tokens, with adaptive thinking always on.

Benchmark gains come with a standard error caveat

On Terminal-Bench-Science 0.1, an agentic scientific research benchmark, Fable 5.1 scores 52.6% against 29.0% for Opus 5, 24.7% for Fable 5, and 22.4% for GPT-5.6 Sol. Anthropic reports a standard error of 3.5 to 4.5 points per model, so the margin between models matters more than the exact ranking. On Terminal-Bench-4.0, Fable 5.1 reaches 55.8% and Mythos 5.1 reaches 60.9%. The gap between the two identical models is the cost of safeguard interventions, which Anthropic disclosed transparently. Other benchmarks include CursorBench 3.2.0 at 73.4%, Humanity's Last Exam at 60.9% without tools and 65.0% with tools, AutomationBench at 31.4%, OSWorld 2.0 at 41.7% strict, and GDPval-AA v2 at 1853.

The Mythos 5.1 variant stays restricted

Claude Mythos 5.1 is the same model as Fable 5.1 but with fewer cyber and life-science safeguards. It remains limited to vetted US organizations under Project Glasswing and is not generally available. This gated access means enterprise pilots and early research access require Anthropic approval.

What changed in safeguards and model behavior

Anthropic documented several regressions and behavior changes. Parallel tool calling is more variable, so agent loops may issue one call per turn where Fable 5 batched several. The model narrates less and answers from memory more often at low effort, and prefers whole-file rewrites over targeted edits. Cyber safeguards now permit vulnerability discovery but not exploit development, cutting interventions in Claude Code by roughly 60% per session. Biology safeguards fire 85% less often on benign requests. Penetration testing, exploit generation, and binary-based vulnerability scanning still redirect to Opus. Content provenance is not optional: a statistical text watermark is applied to all output and C2PA credentials are added to files.

Practical implications for AI builders

For teams building agentic workflows, the 75% cache read reduction directly improves the economics of long-running, context-heavy tasks. The 1M token context window and 128K max output tokens remain unchanged, so workflow boundaries stay the same. The known regressions around parallel tool calling and narration style may require adjustments in agent loop design and user-facing interaction patterns. The safeguard changes mean fewer false positives for security and biology-related queries, but the restrictions on exploit generation remain. The C2PA content provenance requirement may affect compliance workflows for generated content.