Claude Fable 5.1
Anthropic's top tier for the hardest reasoning and long-horizon agent work, which Anthropic recommends when Opus 5.5 still falls short. It is second on Humanity's Last Exam in this list and sits just behind the leaders on the LMArena text leaderboard, but it is also the most expensive model here.
AnthropicLanguage modelReleased September 2026Data checked October 18, 2026
Where people rank it
#1
Chat and everyday questions (tied), 14,590 votes
#2
Writing (tied), 3,284 votes
#2
Web search and research (tied), 12,782 votes
#4
Coding and building apps, 7,503 votes
From blind votes on LMArena, where people pick the better of two anonymous answers. See the full tables on the AI leaderboard.
On paper
Fixed-test rankings from the benchmarks tab: Reasoning #2, Speed #14 (tied), Value #16.
Strengths and weaknesses
- 59.1% on Humanity's Last Exam (Artificial Analysis), second only to Opus 5.5
- Strong long-document reasoning (85.3% on AA-LCR) and 93.7% on GPQA Diamond
- 90.0% on ARC-AGI-2 and a top-five score on the LMArena Code Arena
- Highest token price in this list ($10 / $50 per million) and the highest Artificial Analysis cost per task ($7.63)
- Slower output than Opus 5.5 or Sonnet 5.5 (71 tokens per second on Artificial Analysis)
Best for
- Very hard reasoning problems
- Long research and analysis projects
- Tasks where Opus 5.5 is not enough
Every score we have
Each bar compares this model with the best result any model on this page has on the same test.
- LMArena text arena (human votes)1,502.3Arena score 1502.3 (95% CI 1496.2 to 1508.3); 14,590 votes; entry "Claude Fable 5.1 (Max)"; position 6 of 416 on the full board
- LMArena creative writing1,484.7Arena score 1484.7 (95% CI 1473.3 to 1496.2); 3,284 votes; entry "Claude Fable 5.1 (Max)"; position 9 of 414 on the full board
- LMArena WebDev (coding)1,741.4Arena score 1741.4 (95% CI 1732.4 to 1750.5); 7,503 votes; entry "Claude Fable 5.1 (Max)"; position 5 of 144 on the full board
- LMArena vision1,286.4Arena score 1286.4 (95% CI 1276.7 to 1296.1); 4,648 votes; entry "claude-fable-5.1-max"; position 17 of 165 on the full board
- LMArena Agent Arena0.140IPS score 0.1403 (95% CI 0.1204 to 0.1603); 16,374 sessions; entry "Claude Fable 5.1 (Max)"; position 3 of 52 on the full board
- Artificial Analysis Intelligence Index53.453.4 on the Artificial Analysis Intelligence Index v4.3.2 (AA shows 53), setting "Claude Fable 5.1 (max with fallback)"
- Humanity's Last Exam (Artificial Analysis)59.1%59.1% on Humanity's Last Exam (AA, 2,158 text-only questions), setting "Claude Fable 5.1 (max with fallback)"
- GPQA Diamond, graduate-level science93.7%93.7% on GPQA Diamond (AA; no longer part of the AA index since v4.2), setting "Claude Fable 5.1 (max with fallback)"
- Terminal-Bench 4.0 (Artificial Analysis run)52.0%52.0% on Terminal-Bench 4.0 as run by AA (mini-swe-agent harness), setting "Claude Fable 5.1 (max with fallback)"
- Long-context reasoning (AA-LCR)85.3%85.3% on AA-LCR (long-context reasoning), setting "Claude Fable 5.1 (max with fallback)"
- AA-Omniscience, accuracy minus hallucination43.543.5 on the AA-Omniscience Index (knowledge accuracy minus hallucination, range -100 to 100), setting "Claude Fable 5.1 (max with fallback)"
- Output speed, tokens per second71 t/s71 output tokens per second (AA median), row "Claude Fable 5.1 (max with fallback)"
- Cost per Artificial Analysis test task$7.63$7.63 to run one AA Intelligence Index task, row "Claude Fable 5.1 (max with fallback)"
- Artificial Analysis Coding Agent Index62.262.2 on the AA Coding Agent Index v1.5 (DeepSWE v1.1, Terminal-Bench 4.0, SWE-Atlas-QnA), best listed setup "Claude Code - Fable 5.1 (max) (with fallback)"
- Terminal-Bench 4.0 leaderboard57.9%57.9% resolved on Terminal-Bench 4.0 (plus or minus 3.8, 95% CI) with Claude Code at max effort
- ARC-AGI-2 puzzles90.0%90.00% on ARC-AGI-2 (semi-private set), best listed setting "Claude Fable 5.1 (Max)", $4.49 per task
- Humanity's Last Exam (Scale AI)46.5%46.50% on Scale AI's Humanity's Last Exam leaderboard (plus or minus 2.00), entry "Fable 5.1 (xhigh)"
Compare with
- Claude Opus 5.5Anthropic's flagship for long-running agentic coding and knowledge work, and the model Anthropic suggests starting with for most jobs.
- GPT-6 AstraOpenAI's most capable model, aimed at complex reasoning, coding, research and document creation.
- GPT-6.1 SolOpenAI's mid-priced flagship, described by OpenAI as near-Astra performance at a lower cost.
- Muse WonderMeta's newest frontier model for coding, long-running agents and computer use, released on October 12, 2026 through the Meta Model API and Muse Code.