Benchmark Almanac

Game Arena Bargaining

Gemini 4 Carbon 317 leads since October 2026
Not in model cards

Two language models negotiate how to split a pool of items, each with private item values. Each side tries to reach a good deal without giving away its values.

Kaggle published the leaderboard on July 23rd, 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
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 317

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. 317 Gemini 4 Carbon
  2. 274 GPT-5.5
  3. 252 Claude Fable 5.2
  4. 246 GPT-6 Astra
  5. 244 Gemini 4 Argon
  6. 243 Claude Sonnet 5.5
  7. 239 GPT-6.1 Sol
  8. 234 Claude Opus 5.5
  9. 224 Claude Fable 5.1
  10. 220 Grok 4.20
  11. 218 Grok 4.5
  12. 218 GPT-5.6 Sol
  13. 214 Gemini 3.7 Flash
  14. 214 GPT-6.1 Terra
  15. 214 Grok 4.6
  16. 208 Gemini 3.8 Flash
  17. 206 GPT-6 Sol
  18. 188 Gemini 3.1 Pro
  19. 188 Claude Opus 4.7
  20. 187 o3
  21. 186 GPT-5.4
  22. 180 Gemini 3 Flash
  23. 173 Gemini 3.6 Flash
  24. 172 Claude Opus 4.6
  25. 172 Claude Opus 4.8
  26. 163 Grok 4.7
  27. 163 Claude Haiku 5.5
  28. 161 Claude Sonnet 4.5
  29. 156 GPT-5.6 Luna
  30. 152 Claude Opus 5
  31. 143 GPT-5 mini
  32. 142 GPT-6 Luna
  33. 142 Gemini 3.5 Flash
  34. 132 GPT-5.6 Terra
  35. 126 GPT-5.2
  36. 125 Grok 4.3
  37. 82 Claude Sonnet 5
  38. 75 Claude Sonnet 4.6
  39. 0 GPT-5.4 Mini
  1. 1 Gemini 4 Carbon 317
  2. 2 GPT-5.5 274
  3. 3 Claude Fable 5.2 252
  4. 4 GPT-6 Astra 246
  5. 5 Gemini 4 Argon 244
  6. 6 Claude Sonnet 5.5 243
  7. 7 GPT-6.1 Sol 239
  8. 8 Claude Opus 5.5 234
  9. 9 Claude Fable 5.1 224
  10. 10 Grok 4.20 reasoning 220
  11. 11 Grok 4.5 218
  12. 12 GPT-5.6 Sol 218
  13. 13 Gemini 3.7 Flash 214
  14. 14 GPT-6.1 Terra 214
  15. 15 Grok 4.6 214
  16. 16 Gemini 3.8 Flash 208
  17. 17 GPT-6 Sol 206
  18. 18 Gemini 3.1 Pro 188
  19. 19 Claude Opus 4.7 188
  20. 20 o3 187
  21. 21 GPT-5.4 186
  22. 22 Gemini 3 Flash 180
  23. 23 Gemini 3.6 Flash 173
  24. 24 Claude Opus 4.6 172
  25. 25 Claude Opus 4.8 172
  26. 26 Grok 4.7 163
  27. 27 Claude Haiku 5.5 163
  28. 28 Claude Sonnet 4.5 161
  29. 29 GPT-5.6 Luna 156
  30. 30 Claude Opus 5 152
  31. 31 GPT-5 mini 143
  32. 32 GPT-6 Luna 142
  33. 33 Gemini 3.5 Flash 142
  34. 34 GPT-5.6 Terra 132
  35. 35 GPT-5.2 126
  36. 36 Grok 4.3 125
  37. 37 Claude Sonnet 5 82
  38. 38 Claude Sonnet 4.6 75
  39. 39 GPT-5.4 Mini 0

Source: Official Game Arena Bargaining leaderboard

#ModelScoreSettingDateSource
1 Gemini 4 Carbon 317 Oct 2026 Official leaderboard
2 GPT-5.5 274 Apr 2026 Official leaderboard
3 Claude Fable 5.2 252 Nov 2026 Official leaderboard
4 GPT-6 Astra 246 Sep 2026 Official leaderboard
5 Gemini 4 Argon 244 Sep 2026 Official leaderboard
6 Claude Sonnet 5.5 243 Sep 2026 Official leaderboard
7 GPT-6.1 Sol 239 Sep 2026 Official leaderboard
8 Claude Opus 5.5 234 Sep 2026 Official leaderboard
9 Claude Fable 5.1 224 Sep 2026 Official leaderboard
10 Grok 4.20 220 reasoning Feb 2026 Official leaderboard
11 Grok 4.5 218 Jul 2026 Official leaderboard
12 GPT-5.6 Sol 218 Jul 2026 Official leaderboard
13 Gemini 3.7 Flash 214 Aug 2026 Official leaderboard
14 GPT-6.1 Terra 214 Oct 2026 Official leaderboard
15 Grok 4.6 214 Aug 2026 Official leaderboard
16 Gemini 3.8 Flash 208 Sep 2026 Official leaderboard
17 GPT-6 Sol 206 Sep 2026 Official leaderboard
18 Gemini 3.1 Pro 188 Feb 2026 Official leaderboard
19 Claude Opus 4.7 188 Apr 2026 Official leaderboard
20 o3 187 Apr 2025 Official leaderboard
21 GPT-5.4 186 Mar 2026 Official leaderboard
22 Gemini 3 Flash 180 Dec 2025 Official leaderboard
23 Gemini 3.6 Flash 173 Jul 2026 Official leaderboard
24 Claude Opus 4.6 172 Feb 2026 Official leaderboard
25 Claude Opus 4.8 172 May 2026 Official leaderboard
26 Grok 4.7 163 Sep 2026 Official leaderboard
27 Claude Haiku 5.5 163 Oct 2026 Official leaderboard
28 Claude Sonnet 4.5 161 Sep 2025 Official leaderboard
29 GPT-5.6 Luna 156 Jul 2026 Official leaderboard
30 Claude Opus 5 152 Jul 2026 Official leaderboard
31 GPT-5 mini 143 Aug 2025 Official leaderboard
32 GPT-6 Luna 142 Sep 2026 Official leaderboard
33 Gemini 3.5 Flash 142 May 2026 Official leaderboard
34 GPT-5.6 Terra 132 Jul 2026 Official leaderboard
35 GPT-5.2 126 Dec 2025 Official leaderboard
36 Grok 4.3 125 Apr 2026 Official leaderboard
37 Claude Sonnet 5 82 Jun 2026 Official leaderboard
38 Claude Sonnet 4.6 75 Feb 2026 Official leaderboard
39 GPT-5.4 Mini 0 Mar 2026 Official leaderboard

2 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 Google DeepMind
Own leaderboard 2 leaders
Gemini 4 Carbon 317

Source: Official Game Arena Bargaining 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
0 100 200 300 400 2025 2026 2027 0 100 200 300 400 2025 2026 2027

Source: Official Game Arena Bargaining 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%