Claude Fable 5.1 — Intelligence Index record, but 20% more expensive

Anthropic released Claude Fable 5.1, which scored a record 66 points on the Intelligence Index but became 20% more expensive at $3.76 per task due to increased token usage.

Compiled by Sergey KostenchukPublished 2026-09-02Updated 2026-09-18

2026-09-02 Research Anthropic

🤖 Claude Fable 5.1 — Intelligence Index record, but 20% more expensive

Anthropic released Claude Fable 5.1 on September 1; Artificial Analysis immediately evaluated it: 66 points on the Intelligence Index at max effort — an absolute record, higher than Opus 5 (63) and GPT-5.6 Sol (61). Records: 59.1% on Humanity's Last Exam, 91.4% on Terminal-Bench v2.1, 62.0% on SciCode.

🌍 The record costs more: $3.76 per task vs. $3.14 for Fable 5 (+20%) — 1.7x more output tokens; the cache discount (−75%, down to $0.25 per 1M) saved ~$1.40.

👤 For agents and long tasks, this is the strongest model, but check the effort level: low and max differ 11x in token usage (13.1M vs. 143.7M) with scores of 58–66. Available in Claude API, Bedrock, and Google Cloud (claude-fable-5-1).

Source 1: https://artificialanalysis.ai/articles/claude-fable-5-1 Source 2: https://platform.claude.com/docs/en/models/fable-5-1/overview

Sources