Benchmark Almanac

Game Arena Word Art

Gemini 4 Carbon 2,528 leads since October 2026
Not in model cards

Language models take turns drawing a word as ASCII art and guessing what the other model drew.

Kaggle published the leaderboard on August 18th, 2026 with 35 models on November 1st, 2026. Gameplay logs are a public Kaggle dataset. Part of the Kaggle Game Arena suite.

Made by
Kaggle and Google DeepMind
Released
August 2026
Kind
Multi-agent games
Level
Everyday
Scored
Ranked by matches
Answers
Agent task
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
39
Models scored on its own leaderboard

Gemini 4 Carbon leads with 2,528

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.

39 models. Scroll sideways for the rest.

  1. 2,528 Gemini 4 Carbon
  2. 2,447 Gemini 4 Argon
  3. 2,161 GPT-6 Astra
  4. 2,151 Claude Fable 5.2
  5. 2,076 GPT-6.1 Sol
  6. 1,596 GPT-6.1 Terra
  7. 1,575 Claude Opus 5.5
  8. 1,412 GPT-6 Sol
  9. 1,389 Gemini 3.8 Flash
  10. 1,196 Claude Fable 5.1
  11. 1,106 GPT-5.6 Sol
  12. 1,103 Gemini 3.1 Pro
  13. 1,072 Gemini 3.7 Flash
  14. 1,036 GPT-5.5
  15. 791 Claude Opus 5
  16. 677 Gemini 3 Flash
  17. 643 Gemini 3.5 Flash
  18. 621 Gemini 3.6 Flash
  19. 605 GPT-5.4
  20. 604 Claude Sonnet 5.5
  21. 491 Grok 4.7
  22. 453 Claude Opus 4.6
  23. 370 Claude Opus 4.8
  24. 331 Claude Opus 4.7
  25. 331 Grok 4.6
  26. 295 GPT-5.6 Terra
  27. 277 o3
  28. 259 GPT-5.6 Luna
  29. 254 GPT-6 Luna
  30. 243 GPT-5.2
  31. 218 Claude Sonnet 4.6
  32. 217 Claude Sonnet 4.5
  33. 177 Grok 4.5
  34. 146 Claude Sonnet 5
  35. 126 GPT-5.4 Mini
  36. 102 Claude Haiku 5.5
  37. 67 Grok 4.20
  38. 2 Grok 4.3
  39. 0 GPT-5 mini
  1. 1 Gemini 4 Carbon 2,528
  2. 2 Gemini 4 Argon 2,447
  3. 3 GPT-6 Astra 2,161
  4. 4 Claude Fable 5.2 2,151
  5. 5 GPT-6.1 Sol 2,076
  6. 6 GPT-6.1 Terra 1,596
  7. 7 Claude Opus 5.5 1,575
  8. 8 GPT-6 Sol 1,412
  9. 9 Gemini 3.8 Flash 1,389
  10. 10 Claude Fable 5.1 1,196
  11. 11 GPT-5.6 Sol 1,106
  12. 12 Gemini 3.1 Pro 1,103
  13. 13 Gemini 3.7 Flash 1,072
  14. 14 GPT-5.5 1,036
  15. 15 Claude Opus 5 791
  16. 16 Gemini 3 Flash 677
  17. 17 Gemini 3.5 Flash 643
  18. 18 Gemini 3.6 Flash 621
  19. 19 GPT-5.4 605
  20. 20 Claude Sonnet 5.5 604
  21. 21 Grok 4.7 491
  22. 22 Claude Opus 4.6 453
  23. 23 Claude Opus 4.8 370
  24. 24 Claude Opus 4.7 331
  25. 25 Grok 4.6 331
  26. 26 GPT-5.6 Terra 295
  27. 27 o3 277
  28. 28 GPT-5.6 Luna 259
  29. 29 GPT-6 Luna 254
  30. 30 GPT-5.2 243
  31. 31 Claude Sonnet 4.6 218
  32. 32 Claude Sonnet 4.5 217
  33. 33 Grok 4.5 177
  34. 34 Claude Sonnet 5 146
  35. 35 GPT-5.4 Mini 126
  36. 36 Claude Haiku 5.5 102
  37. 37 Grok 4.20 reasoning 67
  38. 38 Grok 4.3 2
  39. 39 GPT-5 mini 0

Source: Official Game Arena Word Art leaderboard

#ModelScoreSettingDateSource
1 Gemini 4 Carbon 2,528 Oct 2026 Official leaderboard
2 Gemini 4 Argon 2,447 Sep 2026 Official leaderboard
3 GPT-6 Astra 2,161 Sep 2026 Official leaderboard
4 Claude Fable 5.2 2,151 Nov 2026 Official leaderboard
5 GPT-6.1 Sol 2,076 Sep 2026 Official leaderboard
6 GPT-6.1 Terra 1,596 Oct 2026 Official leaderboard
7 Claude Opus 5.5 1,575 Sep 2026 Official leaderboard
8 GPT-6 Sol 1,412 Sep 2026 Official leaderboard
9 Gemini 3.8 Flash 1,389 Sep 2026 Official leaderboard
10 Claude Fable 5.1 1,196 Sep 2026 Official leaderboard
11 GPT-5.6 Sol 1,106 Jul 2026 Official leaderboard
12 Gemini 3.1 Pro 1,103 Feb 2026 Official leaderboard
13 Gemini 3.7 Flash 1,072 Aug 2026 Official leaderboard
14 GPT-5.5 1,036 Apr 2026 Official leaderboard
15 Claude Opus 5 791 Jul 2026 Official leaderboard
16 Gemini 3 Flash 677 Dec 2025 Official leaderboard
17 Gemini 3.5 Flash 643 May 2026 Official leaderboard
18 Gemini 3.6 Flash 621 Jul 2026 Official leaderboard
19 GPT-5.4 605 Mar 2026 Official leaderboard
20 Claude Sonnet 5.5 604 Sep 2026 Official leaderboard
21 Grok 4.7 491 Sep 2026 Official leaderboard
22 Claude Opus 4.6 453 Feb 2026 Official leaderboard
23 Claude Opus 4.8 370 May 2026 Official leaderboard
24 Claude Opus 4.7 331 Apr 2026 Official leaderboard
25 Grok 4.6 331 Aug 2026 Official leaderboard
26 GPT-5.6 Terra 295 Jul 2026 Official leaderboard
27 o3 277 Apr 2025 Official leaderboard
28 GPT-5.6 Luna 259 Jul 2026 Official leaderboard
29 GPT-6 Luna 254 Sep 2026 Official leaderboard
30 GPT-5.2 243 Dec 2025 Official leaderboard
31 Claude Sonnet 4.6 218 Feb 2026 Official leaderboard
32 Claude Sonnet 4.5 217 Sep 2025 Official leaderboard
33 Grok 4.5 177 Jul 2026 Official leaderboard
34 Claude Sonnet 5 146 Jun 2026 Official leaderboard
35 GPT-5.4 Mini 126 Mar 2026 Official leaderboard
36 Claude Haiku 5.5 102 Oct 2026 Official leaderboard
37 Grok 4.20 67 reasoning Feb 2026 Official leaderboard
38 Grok 4.3 2 Apr 2026 Official leaderboard
39 GPT-5 mini 0 Aug 2025 Official leaderboard

6 models have held number one on the official leaderboard since its release in August 2026

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

OpenAI Anthropic Google DeepMind
Own leaderboard 6 leaders
Gemini 4 Carbon 2,528

Source: Official Game Arena Word Art 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
0 1,000 2,000 3,000 2025 2026 2027 0 1,000 2,000 3,000 2025 2026 2027

Source: Official Game Arena Word Art 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 Games and long tasks

All 82 in Games and long tasks
  • ARC-AGI-3 In use Interactive game environments where agents must explore, infer goals and rules, and act efficiently without instructions. GPT-6 Astra 99.95%
  • Orak Fading LLM agents play 12 popular video games across most genres through an MCP interface. Each game scores in-game progress or outcomes, and the board ranks models by their average rank across the games. Gemini 2.5 Pro 3.90
  • MineDojo Fading Agents complete open-ended tasks in Minecraft, such as mining blocks and fighting mobs, from a large simulation suite. Programmatic checks score many tasks. Seed2.0 Pro 49%
  • TextQuests Only in system cards Agents play 25 classic interactive fiction games. The score is harmful actions per 100 steps, where lower is better. The board also reports game progress. Llama 4 Scout 7.00
  • GameWorld Only in system cards Multimodal game agents play 34 browser games across 170 tasks. Outcomes are checked from the game state. GPT-6.1 Terra 76.8%
  • Swarm Bench Only in system cards Models orchestrate agent swarms on complex tasks that benefit from coordinated decomposition and parallel execution. Kimi K3 76.3%