Benchmark Almanac

Agent Island

GPT-5.5 0.88 leads since May 2026
Not in model cards

Models play multi-agent games that require cooperation, conflict, and persuasion.

The May 2026 paper proposes a contamination-resistant multi-agent game benchmark. The abstract does not state the best score.

Released
May 2026
Kind
Multi-agent games
Level
Everyday
Scored
Settled by events
Answers
Agent task
Tasks
All public
Measures
Capability
Language
English
In and out
Text → Text, Actions
Interaction
Multi-agent
Environment
Agent Island
Items from
Live play or votes
Refresh
Live
Run by
Anyone
In views
Games, Agents, Kept fresh
Strongest model tested
Claude Fable 5 #12 on the capability index
Last model added
June 2026
Scores refreshed
November 20th, 2026
58
Models scored on its own leaderboard

GPT-5.5 leads with 0.88 among models released by June 2026

Best score per model from the benchmark's own leaderboard, read on November 20th, 2026. Where a model ran with several settings or harnesses, we keep its best.

The newest model with a score here came out in June 2026, and nobody has tested a newer one. Among the 20 highest-ranked models on the capability index, these have no score yet:

  1. 1 GPT-5.5 0.88
  2. 2 GPT-5.2 0.52
  3. 3 GPT-5.3 Codex 0.49
  4. 4 Claude Fable 5 0.46
  5. 5 GPT-5.4 0.38
  6. 6 Claude Opus 4.7 0.32
  7. 7 Claude Opus 4.6 0.31
  8. 8 Kimi K2.6 0.29
  9. 9 DeepSeek-V4-Pro 0.28
  10. 10 Claude Opus 4.5 0.27
  11. 11 GLM-5.1 0.25
  12. 12 Claude Sonnet 4.6 0.24
  13. 13 o3 0.21
  14. 14 MiniMax-M3 0.19
  15. 15 GPT-5.3 Instant 0.18
  16. 16 Gemini 3.1 Pro 0.17
  17. 17 Kimi K2 Thinking 0.16
  18. 18 GLM-5 0.16
  19. 19 Claude Sonnet 4.5 0.14
  20. 20 Grok 4.20 0.14
  21. 21 Qwen3.7-Max 0.13
  22. 22 GLM-5.2 0.13
  23. 23 Claude Opus 4.8 0.13
  24. 24 Qwen3.5 397B-A17B 0.12
  25. 25 Gemini 2.5 Pro Jun 2025 0.12
  26. 26 Kimi K2.7 Code 0.12
  27. 27 Kimi K2.5 0.12
  28. 28 Grok 4.20 Multi-agent 0.11
  29. 29 Claude Sonnet 5 0.10
  30. 30 GPT-5.4 Mini 0.10
  31. 31 DeepSeek-V3.2 0.09
  32. 32 Claude Haiku 4.5 0.09
  33. 33 Grok 4 0.09
  34. 34 Claude 3.7 Sonnet 0.09
  35. 35 Nemotron 3 Ultra 0.08
  36. 36 Qwen3.6 27B 0.07
  37. 37 Grok 3 0.07
  38. 38 Grok 4 Fast 0.07
  39. 39 Grok-3 mini 0.06
  40. 40 Gemini 3.5 Flash 0.06
  41. 41 MiniMax-M2.5 0.06
  42. 42 Gemini 3 Flash 0.05
  43. 43 o3-mini 0.05
  44. 44 Cohere Command A 0.05
  45. 45 Mistral Medium 3.1 0.05
  46. 46 Gemini 2.5 Flash Sep 2025 0.04
  47. 47 Qwen3.7-Plus 0.04
  48. 48 DeepSeek-R1 May 2025 0.03
  49. 49 Llama 4 Maverick 0.03
  50. 50 Grok 4.3 0.02
  51. 51 Grok 4.1 Fast 0.02
  52. 52 MiniMax-M2.7 0.02
  53. 53 Qwen3.5-Plus 0.01
  54. 54 Kimi K2 Sep 2025 0.01
  55. 55 Llama 4 Scout 0.01
  56. 56 Qwen3-235B-A22B-Thinking Jul 2025 0.01
  57. 57 Command R+ 0.01
  58. 58 o4-mini 0.01

Source: Official Agent Island leaderboard

#ModelScore ±SettingReleasedRun onSource
1 GPT-5.5 0.88 0.7 to 1.4 Apr 2026 Jul 2026 Official leaderboard
2 GPT-5.2 0.52 0.4 to 0.9 Dec 2025 Jul 2026 Official leaderboard
3 GPT-5.3 Codex 0.49 0.4 to 0.8 Feb 2026 Jul 2026 Official leaderboard
4 Claude Fable 5 0.46 0.3 to 1.2 Jun 2026 Jul 2026 Official leaderboard
5 GPT-5.4 0.38 0.3 to 0.6 Mar 2026 Jul 2026 Official leaderboard
6 Claude Opus 4.7 0.32 0.2 to 0.5 Apr 2026 Jul 2026 Official leaderboard
7 Claude Opus 4.6 0.31 0.2 to 0.5 Feb 2026 Jul 2026 Official leaderboard
8 Kimi K2.6 0.29 0.2 to 0.7 Apr 2026 Jul 2026 Official leaderboard
9 DeepSeek-V4-Pro 0.28 0.2 to 0.5 Apr 2026 Jul 2026 Official leaderboard
10 Claude Opus 4.5 0.27 0.2 to 0.4 Nov 2025 Jul 2026 Official leaderboard
11 GLM-5.1 0.25 0.2 to 0.4 Apr 2026 Jul 2026 Official leaderboard
12 Claude Sonnet 4.6 0.24 0.2 to 0.4 Feb 2026 Jul 2026 Official leaderboard
13 o3 0.21 0.1 to 0.5 Apr 2025 Jul 2026 Official leaderboard
14 MiniMax-M3 0.19 0.1 to 0.5 Jun 2026 Jul 2026 Official leaderboard
15 GPT-5.3 Instant 0.18 0.1 to 0.3 Mar 2026 Jul 2026 Official leaderboard
16 Gemini 3.1 Pro 0.17 0.1 to 0.3 Feb 2026 Jul 2026 Official leaderboard
17 Kimi K2 Thinking 0.16 0.1 to 0.3 Nov 2025 Jul 2026 Official leaderboard
18 GLM-5 0.16 0.1 to 0.3 Feb 2026 Jul 2026 Official leaderboard
19 Claude Sonnet 4.5 0.14 0.1 to 0.2 Sep 2025 Jul 2026 Official leaderboard
20 Grok 4.20 0.14 0.1 to 0.2 Feb 2026 Jul 2026 Official leaderboard
21 Qwen3.7-Max 0.13 0.1 to 0.3 May 2026 Jul 2026 Official leaderboard
22 GLM-5.2 0.13 0.1 to 0.5 Jun 2026 Jul 2026 Official leaderboard
23 Claude Opus 4.8 0.13 0.1 to 0.4 May 2026 Jul 2026 Official leaderboard
24 Qwen3.5 397B-A17B 0.12 0.1 to 0.2 Feb 2026 Jul 2026 Official leaderboard
25 Gemini 2.5 Pro Jun 2025 0.12 0.1 to 0.2 Jun 2025 Jul 2026 Official leaderboard
26 Kimi K2.7 Code 0.12 0 to 0.6 Jun 2026 Jul 2026 Official leaderboard
27 Kimi K2.5 0.12 0.1 to 0.2 Jan 2026 Jul 2026 Official leaderboard
28 Grok 4.20 Multi-agent 0.11 0 to 0.4 Mar 2026 Jul 2026 Official leaderboard
29 Claude Sonnet 5 0.10 0 to 0.5 Jun 2026 Jul 2026 Official leaderboard
30 GPT-5.4 Mini 0.10 0.1 to 0.2 Mar 2026 Jul 2026 Official leaderboard
31 DeepSeek-V3.2 0.09 0.1 to 0.2 Dec 2025 Jul 2026 Official leaderboard
32 Claude Haiku 4.5 0.09 0 to 0.3 Oct 2025 Jul 2026 Official leaderboard
33 Grok 4 0.09 0.1 to 0.2 Jul 2025 Jul 2026 Official leaderboard
34 Claude 3.7 Sonnet 0.09 0.1 to 0.2 Feb 2025 Jul 2026 Official leaderboard
35 Nemotron 3 Ultra 0.08 0 to 0.2 Jun 2026 Jul 2026 Official leaderboard
36 Qwen3.6 27B 0.07 0 to 0.2 Apr 2026 Jul 2026 Official leaderboard
37 Grok 3 0.07 0 to 0.2 Apr 2025 Jul 2026 Official leaderboard
38 Grok 4 Fast 0.07 0 to 0.1 Sep 2025 Jul 2026 Official leaderboard
39 Grok-3 mini 0.06 0 to 0.2 Apr 2025 Jul 2026 Official leaderboard
40 Gemini 3.5 Flash 0.06 0 to 0.2 May 2026 Jul 2026 Official leaderboard
41 MiniMax-M2.5 0.06 0 to 0.2 Feb 2026 Jul 2026 Official leaderboard
42 Gemini 3 Flash 0.05 0 to 0.2 Dec 2025 Jul 2026 Official leaderboard
43 o3-mini 0.05 0 to 0.2 Jan 2025 Jul 2026 Official leaderboard
44 Cohere Command A 0.05 0 to 0.2 Mar 2025 Jul 2026 Official leaderboard
45 Mistral Medium 3.1 0.05 0 to 0.2 Aug 2025 Jul 2026 Official leaderboard
46 Gemini 2.5 Flash Sep 2025 0.04 0 to 0.2 Sep 2025 Jul 2026 Official leaderboard
47 Qwen3.7-Plus 0.04 0 to 0.2 Jun 2026 Jul 2026 Official leaderboard
48 DeepSeek-R1 May 2025 0.03 0 to 0.1 May 2025 Jul 2026 Official leaderboard
49 Llama 4 Maverick 0.03 0 to 0.2 Apr 2025 Jul 2026 Official leaderboard
50 Grok 4.3 0.02 0 to 0.1 Apr 2026 Jul 2026 Official leaderboard
51 Grok 4.1 Fast 0.02 0 to 0.1 Nov 2025 Jul 2026 Official leaderboard
52 MiniMax-M2.7 0.02 0 to 0.1 Mar 2026 Jul 2026 Official leaderboard
53 Qwen3.5-Plus 0.01 0 to 0.2 Feb 2026 Jul 2026 Official leaderboard
54 Kimi K2 Sep 2025 0.01 0 to 0.2 Sep 2025 Jul 2026 Official leaderboard
55 Llama 4 Scout 0.01 0 to 0.2 Apr 2025 Jul 2026 Official leaderboard
56 Qwen3-235B-A22B-Thinking Jul 2025 0.01 0 to 0.1 Jul 2025 Jul 2026 Official leaderboard
57 Command R+ 0.01 0 to 0.1 Aug 2024 Jul 2026 Official leaderboard
58 o4-mini 0.01 0 to 0.1 Apr 2025 Jul 2026 Official leaderboard

Number one over time

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

OpenAI
Own leaderboard Not tested since Jun 2026
GPT-5.5 0.88

Source: Official Agent Island leaderboard

Scores over time

All the scores we have, one dot each, so a model tested at several settings appears more than once.

Plot with other benchmarks
On the benchmark’s own leaderboard
0.00 0.20 0.40 0.60 0.80 1.00 2025 2026 0.00 0.20 0.40 0.60 0.80 1.00 2025 2026

Source: Official Agent Island leaderboard

No lab model card reports it yet

We read the headline results of 540 launch posts from 57 labs, and none of them list this benchmark.

More in Games and long tasks

All 261 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-6.2 Astra 100%
  • 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%

Data audit

All 9 checks pass. Every benchmark

Catalog record 2 pass
  • Pass Task access All tasks are public.
  • Pass Confidence Checked against a primary source.
Scores 3 pass 2 notes
  • Pass Independent runs Scores from the official leaderboard (58 entries).
  • Pass Models scored 58 models we track.
  • Pass Newest score The newest is from June 2026, for Claude Sonnet 5.
  • Note Frontier models None of the top 10 has a score. The strongest tested is Claude Fable 5 (#12).
  • Note Headroom Scored as a plain number, with no fixed ceiling.
Leaderboard import 3 pass
  • Pass Importer We read the board on agentisland.ai: 59 rows for models we track at the last read.
  • Pass Last read November 20th, 2026.
  • Pass Read worked The last read worked.
Consistency 1 pass 1 note
  • Note Lab claims No model has both a lab number and an independent run.
  • Pass Known issues No published audit found problems.

How we check