Arena Document
- What does it measure?
- Tests how well models read and analyze long, real user documents (PDFs) to answer questions, summarize, or extract information.
- Who evaluates it, and how?
- A user uploads a PDF and makes a request; two anonymous models respond and the user votes for the better one. The votes are combined into a score on the Elo scale, a rating built from head-to-head results.
- Reading the score
- Scores are Elo ratings from head-to-head human votes. Higher is better.
- Related benchmarks
- Arena Image-to-Video|Arena Search|Arena Text Factuality|Arena Video Edit
Top score1515
Models tested36
Scoring basisHigher is better
Last updated2026.11.09
Model release date
Can it pull scattered facts from several long documents into one answer?Long Context Reasoning
Kimi K3🥇
Kimi K3🥇Long Context Reasoning
Kimi K3🥇Can it read tables and reshape them as needed?LiveBench 2.0 Data Analysis
Gemini 4 Argon🥇
Gemini 4 Argon🥇LiveBench 2.0 Data Analysis
Gemini 4 Argon🥇Which AI gives the business answers people prefer most?Arena Business, Management & Finance
Gemini 4 Argon🥇
Gemini 4 Argon🥇Arena Business, Management & Finance
Gemini 4 Argon🥇
Which AI gives the document analysis answers people prefer most?
Arena Document
- 🥇AnthropicClaude Fable 5.11515
- 🥈AnthropicClaude Opus 5Reasoning effort: High1509
- 🥉AnthropicClaude Opus 4.6Reasoning effort: High1508
- #4AnthropicClaude Fable 51497
- #5AnthropicClaude Opus 4.7Reasoning effort: High1496
- #6OpenAIGPT-5.5Reasoning effort: High1484
- #7GoogleGemini 3.8 Flash1482
- #8AnthropicClaude Sonnet 4.61482
- #9OpenAIGPT-5.6 Sol1480
- #10GoogleGemini 3.7 Flash1476
- #11AnthropicClaude Opus 4.8Reasoning effort: High1475
- #12OpenAIGPT-5.41473
- #13MetaMuse Spark 1.31473
- #14OpenAIGPT-5.6 Terra1468
- #15OpenAIGPT-6 Astra1467
- #16AnthropicClaude Opus 4.51466
- #17MetaMuse Spark 1.11463
- #18GoogleGemini 3.5 FlashReasoning effort: High1462
- #19AnthropicClaude Sonnet 51460
- #20GoogleGemini 3.6 Flash1456
- #21OpenAIGPT-5.6 Luna1454
- #22SpaceXSIGrok 4.51451
- #23Moonshot AIKimi K2.61450
- #24AnthropicClaude Sonnet 4.51449
- #25MetaMuse Spark1444
- #26GoogleGemini 3.1 Pro1443
- #27SpaceXSIGrok 4.61438
- #28AlibabaQwen3.7 Plus1437
- #29MiniMaxMiniMax M31437
- #30GoogleGemma 4 31B1434
- #31Moonshot AIKimi K2.51431
- #32GoogleGemini 2.5 Pro1423
- #33AnthropicClaude Haiku 4.51420
- #34Z.aiGLM 5V Turbo1419
- #35SpaceXSIGrok 4.20 (Reasoning)1417
- #36GoogleGemini 3 Flash1414
Top 5 within the margin of error · effectively tied · Half of models ≤ 1461
The "harness" is the agent program the AI used to carry out the task. The same model can score very differently depending on its harness and reasoning effort.