The arena · AI model review
Claude Fable 5.1
by Anthropic
Anthropic September 2026 Fable refresh: 52.6% Terminal-Bench-Science, 75% cheaper cache reads, same $10/$50
$50/1M out
Official Anthropic API price for claude-fable-5.1: $10/1M input, $50/1M output, cache writes $12.50/MTok (5-min) and $20/MTok (1-hour), cache reads $0.25/MTok (down from $1). 50% batch discount applies. Verified against claude.com/pricing and llm-stats.com 09/2026.
Anthropic
1M tokens
$10/1M in
$50/1M out
text, vision
No
What is Claude Fable 5.1?
Released September 1, 2026 as direct successor to Claude Fable 5. API ID claude-fable-5.1: $10/$50 per 1M tokens (cache reads dropped from $1 to $0.25), 1M context window, 128K max output, always-on adaptive thinking. Headline: 52.6% on Terminal-Bench-Science 0.1 vs 24.7% for Fable 5. Agentic coding: 55.8% Terminal-Bench 4.0, 73.4% CursorBench 3.2.0. Knowledge cutoff June 2026.
Claude Fable 5.1 pros & cons
Pros
- 52.6% on Terminal-Bench-Science 0.1 vs 24.7% for Fable 5 and 29.0% for Opus 5, a 2x jump in scientific agentic capability
- Cache reads dropped 75% from $1 to $0.25 per million, roughly 25% lower cost on typical workloads and up to 45% on agentic ones
- #1 on Vals Index (67.87%) ahead of Opus 5 (67.21%) and Fable 5 (66.04%), 100% on ProofBench v1.1
- #1 on LiveCodeBench (90.52%), #2 on Terminal-Bench 2.1 (85.02%), 90.26% on Vibe Code Bench
- Same $10/$50 pricing as Fable 5, 1M context, 128K output, adaptive thinking on by default
- Available on Anthropic API, Azure, Amazon Bedrock, and Google Vertex from day one
Cons
- Still $10/$50 per million, 2x Opus 5 pricing despite similar scores on some benchmarks
- P95 time to first token of 16.57 seconds, notably slow for interactive applications
- Dual-use safety classifiers can trigger on legitimate scientific and medical content
- Proprietary model with Anthropic usage policy restrictions
- Adaptive thinking always on adds latency and cost on simple tasks
The arena’s verdict
Claude Fable 5.1 is the new ceiling for agentic scientific work. The 2x jump on Terminal-Bench-Science over Fable 5 is real, and the 75% cache-read discount makes long-running agents meaningfully cheaper. Choose it for the hardest research and coding tasks where you need the best published scores and can tolerate 16-second first-token latency. If you do not need the science bump, Opus 5 at $5/$25 delivers similar everyday coding at half the price. For budget-sensitive volume, Sonnet 5 or Gemini Flash remain better fits.
Thumbs up or thumbs down
Cast your verdict
Would you recommend Claude Fable 5.1, or warn the crowd away?
Top Claude Fable 5.1 alternatives
All alternativesGoogle workhorse September 2026: 1M context, $0.75/$3.75 promo pricing, 305 tokens/s output speed
Google's Nov 2025 frontier model: 1M context, first past 1500 Elo on LMArena, replaced by Gemini 3.1 Pro.
Compare Claude Fable 5.1 head-to-head
Claude Fable 5.1: frequently asked questions
How much does Claude Fable 5.1 cost per 1M tokens?
Claude Fable 5.1 costs $10/1M in per 1M input tokens and $50/1M out per 1M output tokens. Official Anthropic API price for claude-fable-5.1: $10/1M input, $50/1M output, cache writes $12.50/MTok (5-min) and $20/MTok (1-hour), cache reads $0.25/MTok (down from $1). 50% batch discount applies. Verified against claude.com/pricing and llm-stats.com 09/2026.