Benchmark Almanac

Document Arena

Claude Fable 5 1,499 leads since Jun 2026 Not in model cards

People upload their own PDFs with a question, compare answers from two anonymous models, and vote for the better one. Votes become Bradley-Terry ratings.

PDF uploads opened in February 2026 and the ranked board went live on 2026-03-03. It had 44 models and 404,715 votes on the 2026-09-13 update. A style-controlled view also exists.

Made by Arena (formerly LMArena) Released Mar 2026 Reasoning over long documents Last model added Sep 2026 Scores refreshed 2 Nov 2026
Level
Everyday Most adults, without training
Scored
Ranked by votes People pick the better of two answers, as in an arena, and the votes are turned into ratings.
Answers
Open-ended Free-form text, documents, designs or media judged as a whole.
Dataset Leaderboard arena.ai
38
Entries on its own leaderboard, 37 models we track

Claude Fable 5 leads with 1,499 among models released by Sep 2026

Best score per entry from the benchmark's own leaderboard, read on 2 Nov 2026. Where a model ran with several settings or harnesses, we keep its best. Every run is in the next chart. Names in gray aren't in our model list: fine-tunes, agent systems, products and older models. They don't count toward leaders.

The newest model with a score here came out in Sep 2026, and nobody has tested a newer one. Among the 20 highest-ranked models on the capability index, these have no score yet:

38 entries. Scroll sideways for the rest. Bars start at 1,400.

  1. 1,499 Claude Fable 5
  2. 1,495 Claude Opus 4.6
  3. 1,494 Claude Opus 4.7
  4. 1,493 Claude Fable 5.1
  5. 1,490 Claude Opus 5
  6. 1,489 GPT-5.6 Sol
  7. 1,486 Claude Opus 4.8
  8. 1,483 GPT-5.5
  9. 1,479 Muse Spark 1.1
  10. 1,478 Claude Sonnet 4.6
  11. 1,477 GPT-5.6 Terra
  12. 1,476 Claude Sonnet 5
  13. 1,473 GPT-6 Astra
  14. 1,470 GPT-5.4
  15. 1,468 Muse Spark 1.3
  16. 1,467 Muse Spark
  17. 1,465 Claude Opus 4.5
  18. 1,464 Grok 4.5
  19. 1,462 GPT-5.6 Luna
  20. 1,461 Grok 4.6
  21. 1,461 Gemini 3.5 Flash
  22. 1,460 GPT-5.5 Instant
  23. 1,459 Gemini 3.1 Pro
  24. 1,453 Gemini 3.6 Flash
  25. 1,453 Qwen3.7-Plus
  26. 1,451 Gemini 3 Pro
  27. 1,450 Kimi K2.6
  28. 1,449 Claude Sonnet 4.5
  29. 1,445 Gemma 4 31B IT
  30. 1,439 grok-4.20-beta-0309-reasoning
  31. 1,439 Gemini 3 Flash
  32. 1,436 Kimi K2.5
  33. 1,433 MiniMax-M3
  34. 1,431 Gemini 2.5 Pro
  35. 1,425 GPT-5.2
  36. 1,419 Claude Haiku 4.5
  37. 1,416 GPT-5.1
  38. 1,400 GLM-5V-Turbo

Bars start at 1,400.

  1. 1 Claude Fable 5 1,499
  2. 2 Claude Opus 4.6 high 1,495
  3. 3 Claude Opus 4.7 1,494
  4. 4 Claude Fable 5.1 max 1,493
  5. 5 Claude Opus 5 high 1,490
  6. 6 GPT-5.6 Sol xhigh 1,489
  7. 7 Claude Opus 4.8 high 1,486
  8. 8 GPT-5.5 1,483
  9. 9 Muse Spark 1.1 1,479
  10. 10 Claude Sonnet 4.6 1,478
  11. 11 GPT-5.6 Terra xhigh 1,477
  12. 12 Claude Sonnet 5 high 1,476
  13. 13 GPT-6 Astra max 1,473
  14. 14 GPT-5.4 1,470
  15. 15 Muse Spark 1.3 max 1,468
  16. 16 Muse Spark 1,467
  17. 17 Claude Opus 4.5 1,465
  18. 18 Grok 4.5 1,464
  19. 19 GPT-5.6 Luna xhigh 1,462
  20. 20 Grok 4.6 high 1,461
  21. 21 Gemini 3.5 Flash medium 1,461
  22. 22 GPT-5.5 Instant 1,460
  23. 23 Gemini 3.1 Pro 1,459
  24. 24 Gemini 3.6 Flash high 1,453
  25. 25 Qwen3.7-Plus 1,453
  26. 26 Gemini 3 Pro 1,451
  27. 27 Kimi K2.6 1,450
  28. 28 Claude Sonnet 4.5 1,449
  29. 29 Gemma 4 31B IT 1,445
  30. 30 grok-4.20-beta-0309-reasoning 1,439
  31. 31 Gemini 3 Flash 1,439
  32. 32 Kimi K2.5 thinking 1,436
  33. 33 MiniMax-M3 1,433
  34. 34 Gemini 2.5 Pro Jun 2025 1,431
  35. 35 GPT-5.2 1,425
  36. 36 Claude Haiku 4.5 1,419
  37. 37 GPT-5.1 1,416
  38. 38 GLM-5V-Turbo 1,400
#EntryScoreSettingDateSource
1 Claude Fable 5 1,499 Jun 2026 Official leaderboard
2 Claude Opus 4.6 1,495 high Feb 2026 Official leaderboard
3 Claude Opus 4.7 1,494 Apr 2026 Official leaderboard
4 Claude Fable 5.1 1,493 max Sep 2026 Official leaderboard
5 Claude Opus 5 1,490 high Jul 2026 Official leaderboard
6 GPT-5.6 Sol 1,489 xhigh Jul 2026 Official leaderboard
7 Claude Opus 4.8 1,486 high May 2026 Official leaderboard
8 GPT-5.5 1,483 Apr 2026 Official leaderboard
9 Muse Spark 1.1 1,479 Jul 2026 Official leaderboard
10 Claude Sonnet 4.6 1,478 Feb 2026 Official leaderboard
11 GPT-5.6 Terra 1,477 xhigh Jul 2026 Official leaderboard
12 Claude Sonnet 5 1,476 high Jun 2026 Official leaderboard
13 GPT-6 Astra 1,473 max Sep 2026 Official leaderboard
14 GPT-5.4 1,470 Mar 2026 Official leaderboard
15 Muse Spark 1.3 1,468 max Sep 2026 Official leaderboard
16 Muse Spark 1,467 Apr 2026 Official leaderboard
17 Claude Opus 4.5 1,465 Nov 2025 Official leaderboard
18 Grok 4.5 1,464 Jul 2026 Official leaderboard
19 GPT-5.6 Luna 1,462 xhigh Jul 2026 Official leaderboard
20 Grok 4.6 1,461 high Aug 2026 Official leaderboard
21 Gemini 3.5 Flash 1,461 medium May 2026 Official leaderboard
22 GPT-5.5 Instant 1,460 May 2026 Official leaderboard
23 Gemini 3.1 Pro 1,459 Feb 2026 Official leaderboard
24 Gemini 3.6 Flash 1,453 high Jul 2026 Official leaderboard
25 Qwen3.7-Plus 1,453 Jun 2026 Official leaderboard
26 Gemini 3 Pro 1,451 Nov 2025 Official leaderboard
27 Kimi K2.6 1,450 Apr 2026 Official leaderboard
28 Claude Sonnet 4.5 1,449 Sep 2025 Official leaderboard
29 Gemma 4 31B IT 1,445 Apr 2026 Official leaderboard
30 grok-4.20-beta-0309-reasoning 1,439 Sep 2026 Official leaderboard
31 Gemini 3 Flash 1,439 Dec 2025 Official leaderboard
32 Kimi K2.5 1,436 thinking Jan 2026 Official leaderboard
33 MiniMax-M3 1,433 Jun 2026 Official leaderboard
34 Gemini 2.5 Pro Jun 2025 1,431 Jun 2025 Official leaderboard
35 GPT-5.2 1,425 Dec 2025 Official leaderboard
36 Claude Haiku 4.5 1,419 Oct 2025 Official leaderboard
37 GPT-5.1 1,416 Nov 2025 Official leaderboard
38 GLM-5V-Turbo 1,400 Apr 2026 Official leaderboard

Scores at each effort level

Models the leaderboard ran at two or more effort levels. We don’t have enough cost data here to say what the extra effort costs.

Lowest effort to highest, light to dark
Claude Opus 4.6
high
1,495
Claude Opus 4.7
high
1,494
Claude Opus 4.8
high
1,486
GPT-5.5
high
1,483
Gemini 3.5 Flash
medium high
1,461
GPT-5.2
high
1,425
ModelEffortHarnessScoreCost per taskOutput tokens
Claude Fable 5 1,499
Claude Opus 4.6 high 1,495
Claude Opus 4.7 1,494
Claude Opus 4.6 1,494
Claude Opus 4.7 high 1,493
Claude Fable 5.1 max 1,493
Claude Opus 5 high 1,490
GPT-5.6 Sol xhigh 1,489
Claude Opus 4.8 high 1,486
GPT-5.5 1,483
GPT-5.5 high 1,480
Muse Spark 1.1 1,479
Claude Sonnet 4.6 1,478
GPT-5.6 Terra xhigh 1,477
Claude Sonnet 5 high 1,476
Claude Opus 4.8 1,475
GPT-6 Astra max 1,473
GPT-5.4 1,470
Muse Spark 1.3 max 1,468
Muse Spark 1,467
Claude Opus 4.5 1,465
Grok 4.5 1,464
GPT-5.6 Luna xhigh 1,462
Grok 4.6 high 1,461
Gemini 3.5 Flash medium 1,461
GPT-5.5 Instant 1,460
Gemini 3.1 Pro 1,459
Gemini 3.6 Flash high 1,453
Gemini 3.5 Flash high 1,453
Qwen3.7-Plus 1,453
Gemini 3 Pro 1,451
Kimi K2.6 1,450
Claude Sonnet 4.5 1,449
Gemma 4 31B IT 1,445
Gemini 3 Flash 1,439
Kimi K2.5 thinking 1,436
MiniMax-M3 1,433
Gemini 2.5 Pro Jun 2025 1,431
GPT-5.2 1,425
GPT-5.2 high 1,423
Claude Haiku 4.5 1,419
GPT-5.1 1,416
GLM-5V-Turbo 1,400

2 models have held number one since its release in Mar 2026

Lab claims count from the day the card came out, independent runs from the model’s release.

Anthropic
Own leaderboard Not tested since Sep 2026
Claude Fable 5 1,499

Scores over time

All the scores we have, one dot each, so a model tested at several settings appears more than once.

On the benchmark’s own leaderboard
1,400 1,420 1,440 1,460 1,480 1,500 2025 2026 1,400 1,420 1,440 1,460 1,480 1,500 2025 2026

No lab model card reports it yet

We read the headline results of 479 launch posts from 51 labs, and none of them list this benchmark.

More in Long context

All 66 in Long context
  • MRCR v2 In use Multi-round coreference: retrieve the n-th of several similar requests hidden in a long synthetic conversation, up to 1M tokens. Muse Spark 1.3 98.5%
  • LongBench v2 In use 503 multiple-choice questions over 8K-2M-word contexts (QA, dialogue, code repos, structured data). Accuracy. Beam 1.0 66.9%
  • GraphWalks In use A graph is given as an edge list in long context. The model answers BFS or parent queries, scored by F1. Claude Fable 5 100%
  • OpenAI MRCR v2 In use Models read a long synthetic chat that hides 2, 4 or 8 identical writing requests and must reproduce the response to the i-th one. Scored by mean string match ratio per context bin. K-EXAONE 2.0 94.4%
  • CL-bench In use Model must learn new knowledge from a supplied long context (rules, procedures, data) and solve tasks. Strict LLM-verified rubrics. GPT-5.4 27.9%
  • AA-LCR 1.1 In use Models answer 100 questions that need reasoning across sets of real documents of about 100K tokens. An LLM equality checker with a fixed system prompt compares each answer with the key. Step 5 Preview 88.3%