Benchmark Almanac

Game Arena Dark Hex

Gemini 3.8 Flash 581 leads since September 2026
Not in model cards

Language models play Hex without seeing the opponent's stones, so they must probe cells to learn the hidden board while building a path.

Kaggle published the leaderboard on June 4th, 2026 with 38 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
June 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
41
Models scored on its own leaderboard

Gemini 3.8 Flash leads with 581

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.

41 models. Scroll sideways for the rest.

  1. 581 Gemini 3.8 Flash
  2. 552 Gemini 4 Carbon
  3. 492 GPT-5.5
  4. 477 Gemini 4 Argon
  5. 439 Gemini 3.5 Flash
  6. 387 GPT-5.4
  7. 359 Gemini 3 Flash
  8. 354 GPT-5.6 Sol
  9. 352 Claude Opus 5
  10. 348 GPT-6.1 Terra
  11. 341 Gemini 3.6 Flash
  12. 319 Claude Fable 5.2
  13. 295 Grok 4.6
  14. 294 Gemini 3.1 Pro
  15. 292 Gemini 3.7 Flash
  16. 283 Claude Opus 5.5
  17. 281 Claude Sonnet 5.5
  18. 279 GPT-6 Astra
  19. 271 GPT-6.1 Sol
  20. 265 Claude Opus 4.8
  21. 262 Grok 4.7
  22. 252 Claude Fable 5.1
  23. 250 GPT-6 Sol
  24. 249 GPT-5.2
  25. 241 Claude Haiku 5.5
  26. 223 o3
  27. 206 Claude Sonnet 5
  28. 188 GPT-5 mini
  29. 173 Grok 4.5
  30. 144 GPT-5.6 Luna
  31. 137 GPT-6 Luna
  32. 129 GPT-5.6 Terra
  33. 126 Grok 4.1 Fast
  34. 115 Claude Sonnet 4.6
  35. 108 Claude Opus 4.6
  36. 94 GPT-5.4 Mini
  37. 93 Claude Opus 4.7
  38. 67 Grok 4.3
  39. 44 Grok 4.20
  40. 40 Claude Sonnet 4.5
  41. 0 DeepSeek-V3.2
  1. 1 Gemini 3.8 Flash 581
  2. 2 Gemini 4 Carbon 552
  3. 3 GPT-5.5 492
  4. 4 Gemini 4 Argon 477
  5. 5 Gemini 3.5 Flash 439
  6. 6 GPT-5.4 387
  7. 7 Gemini 3 Flash 359
  8. 8 GPT-5.6 Sol 354
  9. 9 Claude Opus 5 352
  10. 10 GPT-6.1 Terra 348
  11. 11 Gemini 3.6 Flash 341
  12. 12 Claude Fable 5.2 319
  13. 13 Grok 4.6 295
  14. 14 Gemini 3.1 Pro 294
  15. 15 Gemini 3.7 Flash 292
  16. 16 Claude Opus 5.5 283
  17. 17 Claude Sonnet 5.5 281
  18. 18 GPT-6 Astra 279
  19. 19 GPT-6.1 Sol 271
  20. 20 Claude Opus 4.8 265
  21. 21 Grok 4.7 262
  22. 22 Claude Fable 5.1 252
  23. 23 GPT-6 Sol 250
  24. 24 GPT-5.2 249
  25. 25 Claude Haiku 5.5 241
  26. 26 o3 223
  27. 27 Claude Sonnet 5 206
  28. 28 GPT-5 mini 188
  29. 29 Grok 4.5 173
  30. 30 GPT-5.6 Luna 144
  31. 31 GPT-6 Luna 137
  32. 32 GPT-5.6 Terra 129
  33. 33 Grok 4.1 Fast reasoning 126
  34. 34 Claude Sonnet 4.6 115
  35. 35 Claude Opus 4.6 108
  36. 36 GPT-5.4 Mini 94
  37. 37 Claude Opus 4.7 93
  38. 38 Grok 4.3 67
  39. 39 Grok 4.20 reasoning 44
  40. 40 Claude Sonnet 4.5 40
  41. 41 DeepSeek-V3.2 0

Source: Official Game Arena Dark Hex leaderboard

#ModelScoreSettingDateSource
1 Gemini 3.8 Flash 581 Sep 2026 Official leaderboard
2 Gemini 4 Carbon 552 Oct 2026 Official leaderboard
3 GPT-5.5 492 Apr 2026 Official leaderboard
4 Gemini 4 Argon 477 Sep 2026 Official leaderboard
5 Gemini 3.5 Flash 439 May 2026 Official leaderboard
6 GPT-5.4 387 Mar 2026 Official leaderboard
7 Gemini 3 Flash 359 Dec 2025 Official leaderboard
8 GPT-5.6 Sol 354 Jul 2026 Official leaderboard
9 Claude Opus 5 352 Jul 2026 Official leaderboard
10 GPT-6.1 Terra 348 Oct 2026 Official leaderboard
11 Gemini 3.6 Flash 341 Jul 2026 Official leaderboard
12 Claude Fable 5.2 319 Nov 2026 Official leaderboard
13 Grok 4.6 295 Aug 2026 Official leaderboard
14 Gemini 3.1 Pro 294 Feb 2026 Official leaderboard
15 Gemini 3.7 Flash 292 Aug 2026 Official leaderboard
16 Claude Opus 5.5 283 Sep 2026 Official leaderboard
17 Claude Sonnet 5.5 281 Sep 2026 Official leaderboard
18 GPT-6 Astra 279 Sep 2026 Official leaderboard
19 GPT-6.1 Sol 271 Sep 2026 Official leaderboard
20 Claude Opus 4.8 265 May 2026 Official leaderboard
21 Grok 4.7 262 Sep 2026 Official leaderboard
22 Claude Fable 5.1 252 Sep 2026 Official leaderboard
23 GPT-6 Sol 250 Sep 2026 Official leaderboard
24 GPT-5.2 249 Dec 2025 Official leaderboard
25 Claude Haiku 5.5 241 Oct 2026 Official leaderboard
26 o3 223 Apr 2025 Official leaderboard
27 Claude Sonnet 5 206 Jun 2026 Official leaderboard
28 GPT-5 mini 188 Aug 2025 Official leaderboard
29 Grok 4.5 173 Jul 2026 Official leaderboard
30 GPT-5.6 Luna 144 Jul 2026 Official leaderboard
31 GPT-6 Luna 137 Sep 2026 Official leaderboard
32 GPT-5.6 Terra 129 Jul 2026 Official leaderboard
33 Grok 4.1 Fast 126 reasoning Nov 2025 Official leaderboard
34 Claude Sonnet 4.6 115 Feb 2026 Official leaderboard
35 Claude Opus 4.6 108 Feb 2026 Official leaderboard
36 GPT-5.4 Mini 94 Mar 2026 Official leaderboard
37 Claude Opus 4.7 93 Apr 2026 Official leaderboard
38 Grok 4.3 67 Apr 2026 Official leaderboard
39 Grok 4.20 44 reasoning Feb 2026 Official leaderboard
40 Claude Sonnet 4.5 40 Sep 2025 Official leaderboard
41 DeepSeek-V3.2 0 Dec 2025 Official leaderboard

2 models have held number one on the official leaderboard since its release in June 2026

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

OpenAI Google DeepMind
Own leaderboard 2 leaders
Gemini 3.8 Flash 581

Source: Official Game Arena Dark Hex 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 2025 2026 2027 0 200 400 600 2025 2026 2027

Source: Official Game Arena Dark Hex 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%