Benchmark Almanac

Game Arena Ultimate Tic-Tac-Toe

Gemini 4 Carbon 720 leads since October 2026
Not in model cards

Language models play nine linked tic-tac-toe boards. Each move decides which small board the opponent must play next.

Kaggle published the leaderboard on July 28th, 2026 with 36 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
July 2026
Kind
Puzzles and chess
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 720

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. 720 Gemini 4 Carbon
  2. 654 Claude Fable 5.2
  3. 600 GPT-6 Astra
  4. 577 Claude Opus 5
  5. 534 Grok 4.7
  6. 500 Grok 4.6
  7. 495 GPT-6.1 Sol
  8. 481 GPT-5.5
  9. 477 Gemini 4 Argon
  10. 443 GPT-5.6 Sol
  11. 439 Claude Fable 5.1
  12. 421 Claude Opus 4.8
  13. 390 GPT-6 Sol
  14. 380 Gemini 3.8 Flash
  15. 377 Grok 4.5
  16. 366 GPT-5.4
  17. 361 GPT-6.1 Terra
  18. 358 Claude Opus 5.5
  19. 357 Gemini 3.1 Pro
  20. 284 Claude Opus 4.6
  21. 269 Gemini 3.6 Flash
  22. 267 Gemini 3.7 Flash
  23. 261 Gemini 3.5 Flash
  24. 245 Claude Sonnet 5.5
  25. 242 GPT-5.2
  26. 241 o3
  27. 186 Claude Sonnet 5
  28. 186 Claude Sonnet 4.6
  29. 176 Claude Sonnet 4.5
  30. 167 Grok 4.20
  31. 165 Gemini 3 Flash
  32. 86 Claude Haiku 5.5
  33. 64 Claude Opus 4.7
  34. 57 GPT-5.6 Terra
  35. 29 Grok 4.3
  36. 25 GPT-6 Luna
  37. 22 GPT-5.6 Luna
  38. 17 GPT-5 mini
  39. 0 GPT-5.4 Mini
  1. 1 Gemini 4 Carbon 720
  2. 2 Claude Fable 5.2 654
  3. 3 GPT-6 Astra 600
  4. 4 Claude Opus 5 577
  5. 5 Grok 4.7 534
  6. 6 Grok 4.6 500
  7. 7 GPT-6.1 Sol 495
  8. 8 GPT-5.5 481
  9. 9 Gemini 4 Argon 477
  10. 10 GPT-5.6 Sol 443
  11. 11 Claude Fable 5.1 439
  12. 12 Claude Opus 4.8 421
  13. 13 GPT-6 Sol 390
  14. 14 Gemini 3.8 Flash 380
  15. 15 Grok 4.5 377
  16. 16 GPT-5.4 366
  17. 17 GPT-6.1 Terra 361
  18. 18 Claude Opus 5.5 358
  19. 19 Gemini 3.1 Pro 357
  20. 20 Claude Opus 4.6 284
  21. 21 Gemini 3.6 Flash 269
  22. 22 Gemini 3.7 Flash 267
  23. 23 Gemini 3.5 Flash 261
  24. 24 Claude Sonnet 5.5 245
  25. 25 GPT-5.2 242
  26. 26 o3 241
  27. 27 Claude Sonnet 5 186
  28. 28 Claude Sonnet 4.6 186
  29. 29 Claude Sonnet 4.5 176
  30. 30 Grok 4.20 reasoning 167
  31. 31 Gemini 3 Flash 165
  32. 32 Claude Haiku 5.5 86
  33. 33 Claude Opus 4.7 64
  34. 34 GPT-5.6 Terra 57
  35. 35 Grok 4.3 29
  36. 36 GPT-6 Luna 25
  37. 37 GPT-5.6 Luna 22
  38. 38 GPT-5 mini 17
  39. 39 GPT-5.4 Mini 0

Source: Official Game Arena Ultimate Tic-Tac-Toe leaderboard

#ModelScoreSettingDateSource
1 Gemini 4 Carbon 720 Oct 2026 Official leaderboard
2 Claude Fable 5.2 654 Nov 2026 Official leaderboard
3 GPT-6 Astra 600 Sep 2026 Official leaderboard
4 Claude Opus 5 577 Jul 2026 Official leaderboard
5 Grok 4.7 534 Sep 2026 Official leaderboard
6 Grok 4.6 500 Aug 2026 Official leaderboard
7 GPT-6.1 Sol 495 Sep 2026 Official leaderboard
8 GPT-5.5 481 Apr 2026 Official leaderboard
9 Gemini 4 Argon 477 Sep 2026 Official leaderboard
10 GPT-5.6 Sol 443 Jul 2026 Official leaderboard
11 Claude Fable 5.1 439 Sep 2026 Official leaderboard
12 Claude Opus 4.8 421 May 2026 Official leaderboard
13 GPT-6 Sol 390 Sep 2026 Official leaderboard
14 Gemini 3.8 Flash 380 Sep 2026 Official leaderboard
15 Grok 4.5 377 Jul 2026 Official leaderboard
16 GPT-5.4 366 Mar 2026 Official leaderboard
17 GPT-6.1 Terra 361 Oct 2026 Official leaderboard
18 Claude Opus 5.5 358 Sep 2026 Official leaderboard
19 Gemini 3.1 Pro 357 Feb 2026 Official leaderboard
20 Claude Opus 4.6 284 Feb 2026 Official leaderboard
21 Gemini 3.6 Flash 269 Jul 2026 Official leaderboard
22 Gemini 3.7 Flash 267 Aug 2026 Official leaderboard
23 Gemini 3.5 Flash 261 May 2026 Official leaderboard
24 Claude Sonnet 5.5 245 Sep 2026 Official leaderboard
25 GPT-5.2 242 Dec 2025 Official leaderboard
26 o3 241 Apr 2025 Official leaderboard
27 Claude Sonnet 5 186 Jun 2026 Official leaderboard
28 Claude Sonnet 4.6 186 Feb 2026 Official leaderboard
29 Claude Sonnet 4.5 176 Sep 2025 Official leaderboard
30 Grok 4.20 167 reasoning Feb 2026 Official leaderboard
31 Gemini 3 Flash 165 Dec 2025 Official leaderboard
32 Claude Haiku 5.5 86 Oct 2026 Official leaderboard
33 Claude Opus 4.7 64 Apr 2026 Official leaderboard
34 GPT-5.6 Terra 57 Jul 2026 Official leaderboard
35 Grok 4.3 29 Apr 2026 Official leaderboard
36 GPT-6 Luna 25 Sep 2026 Official leaderboard
37 GPT-5.6 Luna 22 Jul 2026 Official leaderboard
38 GPT-5 mini 17 Aug 2025 Official leaderboard
39 GPT-5.4 Mini 0 Mar 2026 Official leaderboard

4 models have held number one on the official leaderboard since its release in July 2026

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

OpenAI Anthropic Google DeepMind
Own leaderboard 4 leaders
Gemini 4 Carbon 720

Source: Official Game Arena Ultimate Tic-Tac-Toe 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
0 200 400 600 800 2025 2026 2027 0 200 400 600 800 2025 2026 2027

Source: Official Game Arena Ultimate Tic-Tac-Toe 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%