Benchmark Almanac

DesignArena PDF

Gemini 4 Carbon 1,335 leads since October 2026
Not in model cards

Agents build a PDF document from a user prompt. Users compare the two documents and vote for the better one.

One of 40 DesignArena boards. DesignArena ranks entries with Bradley-Terry ratings from user votes.

Kind
Documents and spreadsheets
Level
Everyday
Scored
Ranked by votes
Answers
Open-ended
Tasks
All public
Strongest model tested
Claude Fable 5.2 #1 on the capability index
Last model added
November 2026
Scores refreshed
November 12th, 2026
37
Models scored on its own leaderboard

Gemini 4 Carbon leads with 1,335

Best score per model from the benchmark's own leaderboard, read on November 12th, 2026. Where a model ran with several settings or harnesses, we keep its best.

37 models. Scroll sideways for the rest. Bars start at 900.

  1. 1,335 Gemini 4 Carbon
  2. 1,310 Claude Opus 5.5
  3. 1,296 Claude Fable 5.2
  4. 1,287 Qwen 3.8 Max
  5. 1,278 Kimi K3
  6. 1,278 Gemini 4 Argon
  7. 1,260 MiniMax-M3
  8. 1,243 GLM-5.2
  9. 1,239 Grok 4.6
  10. 1,238 Claude Sonnet 5.5
  11. 1,228 Claude Opus 5
  12. 1,228 Kimi K3.1 Pro
  13. 1,204 Claude Sonnet 5
  14. 1,202 Claude Opus 4.6
  15. 1,200 GLM-5.1
  16. 1,200 Muse Wonder
  17. 1,199 Gemini 3.8 Flash
  18. 1,194 GPT-6 Astra
  19. 1,189 Muse Spark 1.3
  20. 1,187 MiniMax-M3.1-Flash
  21. 1,178 Gemini 3.6 Flash
  22. 1,173 GPT-5.6 Sol
  23. 1,168 Grok 4.7
  24. 1,163 Gemini 3.7 Flash
  25. 1,156 Claude Opus 4.8
  26. 1,143 Muse Spark 1.1
  27. 1,142 Qwen3.7-Max
  28. 1,139 GPT-6.1 Sol
  29. 1,137 Grok 4.5
  30. 1,134 Gemini 3.5 Flash
  31. 1,120 GPT-6 Luna
  32. 1,116 Muse Spark 1.2
  33. 1,104 Gemini 3.1 Pro
  34. 1,092 GPT-6 Sol
  35. 1,082 GPT-6.1 Terra
  36. 1,032 GPT-5.6 Terra
  37. 928 Grok 4.3

Bars start at 900.

  1. 1 Gemini 4 Carbon 1,335
  2. 2 Claude Opus 5.5 1,310
  3. 3 Claude Fable 5.2 1,296
  4. 4 Qwen 3.8 Max 1,287
  5. 5 Kimi K3 1,278
  6. 6 Gemini 4 Argon 1,278
  7. 7 MiniMax-M3 1,260
  8. 8 GLM-5.2 1,243
  9. 9 Grok 4.6 1,239
  10. 10 Claude Sonnet 5.5 1,238
  11. 11 Claude Opus 5 1,228
  12. 12 Kimi K3.1 Pro 1,228
  13. 13 Claude Sonnet 5 1,204
  14. 14 Claude Opus 4.6 1,202
  15. 15 GLM-5.1 1,200
  16. 16 Muse Wonder max 1,200
  17. 17 Gemini 3.8 Flash 1,199
  18. 18 GPT-6 Astra xhigh 1,194
  19. 19 Muse Spark 1.3 xhigh, max 1,189
  20. 20 MiniMax-M3.1-Flash 1,187
  21. 21 Gemini 3.6 Flash 1,178
  22. 22 GPT-5.6 Sol medium, xhigh 1,173
  23. 23 Grok 4.7 1,168
  24. 24 Gemini 3.7 Flash 1,163
  25. 25 Claude Opus 4.8 1,156
  26. 26 Muse Spark 1.1 1,143
  27. 27 Qwen3.7-Max 1,142
  28. 28 GPT-6.1 Sol 1,139
  29. 29 Grok 4.5 1,137
  30. 30 Gemini 3.5 Flash 1,134
  31. 31 GPT-6 Luna 1,120
  32. 32 Muse Spark 1.2 1,116
  33. 33 Gemini 3.1 Pro 1,104
  34. 34 GPT-6 Sol 1,092
  35. 35 GPT-6.1 Terra 1,082
  36. 36 GPT-5.6 Terra 1,032
  37. 37 Grok 4.3 928

Source: Official DesignArena PDF leaderboard

#ModelScoreSettingDateSource
1 Gemini 4 Carbon 1,335 Oct 2026 Official leaderboard
2 Claude Opus 5.5 1,310 Sep 2026 Official leaderboard
3 Claude Fable 5.2 1,296 Nov 2026 Official leaderboard
4 Qwen 3.8 Max 1,287 Aug 2026 Official leaderboard
5 Kimi K3 1,278 Jul 2026 Official leaderboard
6 Gemini 4 Argon 1,278 Sep 2026 Official leaderboard
7 MiniMax-M3 1,260 Jun 2026 Official leaderboard
8 GLM-5.2 1,243 Jun 2026 Official leaderboard
9 Grok 4.6 1,239 Aug 2026 Official leaderboard
10 Claude Sonnet 5.5 1,238 Sep 2026 Official leaderboard
11 Claude Opus 5 1,228 Jul 2026 Official leaderboard
12 Kimi K3.1 Pro 1,228 Oct 2026 Official leaderboard
13 Claude Sonnet 5 1,204 Jun 2026 Official leaderboard
14 Claude Opus 4.6 1,202 Feb 2026 Official leaderboard
15 GLM-5.1 1,200 Apr 2026 Official leaderboard
16 Muse Wonder 1,200 max Oct 2026 Official leaderboard
17 Gemini 3.8 Flash 1,199 Sep 2026 Official leaderboard
18 GPT-6 Astra 1,194 xhigh Sep 2026 Official leaderboard
19 Muse Spark 1.3 1,189 xhigh, max Sep 2026 Official leaderboard
20 MiniMax-M3.1-Flash 1,187 Oct 2026 Official leaderboard
21 Gemini 3.6 Flash 1,178 Jul 2026 Official leaderboard
22 GPT-5.6 Sol 1,173 medium, xhigh Jul 2026 Official leaderboard
23 Grok 4.7 1,168 Sep 2026 Official leaderboard
24 Gemini 3.7 Flash 1,163 Aug 2026 Official leaderboard
25 Claude Opus 4.8 1,156 May 2026 Official leaderboard
26 Muse Spark 1.1 1,143 Jul 2026 Official leaderboard
27 Qwen3.7-Max 1,142 May 2026 Official leaderboard
28 GPT-6.1 Sol 1,139 Sep 2026 Official leaderboard
29 Grok 4.5 1,137 Jul 2026 Official leaderboard
30 Gemini 3.5 Flash 1,134 May 2026 Official leaderboard
31 GPT-6 Luna 1,120 Sep 2026 Official leaderboard
32 Muse Spark 1.2 1,116 Aug 2026 Official leaderboard
33 Gemini 3.1 Pro 1,104 Feb 2026 Official leaderboard
34 GPT-6 Sol 1,092 Sep 2026 Official leaderboard
35 GPT-6.1 Terra 1,082 Oct 2026 Official leaderboard
36 GPT-5.6 Terra 1,032 Jul 2026 Official leaderboard
37 Grok 4.3 928 Apr 2026 Official leaderboard

6 models have held number one on the official leaderboard since 2023

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

Anthropic MiniMax Moonshot AI Qwen Google DeepMind
Own leaderboard 6 leaders
Gemini 4 Carbon 1,335

Source: Official DesignArena PDF leaderboard

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
900 1,000 1,100 1,200 1,300 1,400 2026 2027 900 1,000 1,100 1,200 1,300 1,400 2026 2027

Source: Official DesignArena PDF leaderboard

No lab model card reports it yet

We read the headline results of 522 launch posts from 52 labs, and none of them list this benchmark.

More in Economic work

All 97 in Economic work
  • JobBench In use 130 tasks from 35 occupations workers most want delegated. Agent works through messy reference files, graded by fact-anchored binary rubrics. GLM-5.3 61.4%
  • GDPval In use Real work tasks from 44 occupations with reference files. Experts blindly compare model deliverables with expert work. Metric is win-or-tie rate. Seed2.1 Pro 87.9%
  • GDPval-AA v2.1 In use Agents with shell and web access complete 220 GDPval gold tasks from 44 occupations. Blind pairwise comparisons give Elo ratings. Claude Opus 5.5 1,846
  • AA-Briefcase v1.1 In use Version 1.1 of Artificial Analysis's multi-hour office work evaluation, with Elo ratings fitted by Crowd-BT. Claude Opus 5.5 1,822
  • CoWorkBench In use Agent completes long-horizon productivity tasks across computer science, finance, law and medicine. Qwen in-house scoring (method unpublished). Claude Fable 5 75.9%
  • OfficeQA In use Grounded reasoning over a large corpus of U.S. Treasury Bulletin documents: locate tables and compute precise numeric answers. Claude Opus 4.7 86.3%