Benchmark Almanac

DuelLab GameBench 2

Claude Fable 5.2 99.1 leads since November 2026
Not in model cards

Models get the rules of a strategy game and write a program that plays it. DuelLab checks and builds each program, then runs head-to-head matches between programs from different models. Match outcomes become per-game scores from 0 to 100 and a Bradley-Terry ranking, with no human or model judge.

GameBench 2 launched on July 14th, 2026 and replaced GameBench 1, a frozen archive whose scores do not compare. Each reasoning setting of a model ranks separately, and a failed program counts as 0. The October 2026 release used eight games. The evaluator and game engine are private, so outsiders cannot rerun every step.

Released
July 2026
Kind
Multi-agent games
Level
University
Scored
Ranked by matches
Answers
Code
Strongest model tested
Claude Fable 5.2 #1 on the capability index
Last model added
November 2026
Scores refreshed
November 12th, 2026
68
Entries on its own leaderboard, 57 models we track

Claude Fable 5.2 leads with 99.1

Best score per entry from the benchmark's own leaderboard, read on November 12th, 2026. Where a model ran with several settings or harnesses, we keep its best. Every run is in the next chart. Names in gray aren't in our model list: fine-tunes, agent systems, products and older models. They don't count toward leaders.

68 entries. Scroll sideways for the rest.

  1. 99.1 Claude Fable 5.2
  2. 94.8 Claude Opus 5.5
  3. 94.3 GPT-6 Astra
  4. 94.0 Claude Fable 5.1
  5. 92.0 GPT-6.1 Sol
  6. 91.4 Claude Sonnet 5.5
  7. 89.7 Pareto 26.10
  8. 89.2 Gemini 4 Argon
  9. 88.3 Pareto 26.10 Preview
  10. 85.8 GPT-6.1 Terra
  11. 85.1 GPT-6.1 Terra Pro
  12. 84.8 Claude Opus 5
  13. 84.0 Kimi K3.1 Pro
  14. 80.1 GPT-5.6 Sol
  15. 79.4 GPT-6 Sol
  16. 77.8 Muse Wonder
  17. 75.2 Claude Fable 5
  18. 75.1 Claude Haiku 5.5
  19. 72.3 GLM-5.3
  20. 72.3 Grok 4.7
  21. 71.7 Particle-1
  22. 70.9 MiMo-V2.6 Pro
  23. 69.9 DeepSeek V4.1 Flash 1017
  24. 69.8 GLM-5.3-FlashX-Legal
  25. 69.7 Qwen3.8-Omni-Flash
  26. 69.6 Qwen3.8 Max (0902)
  27. 68.7 Grok 4.6
  28. 68.5 GLM-5.3 FlashX
  29. 68.5 Muse Spark 1.3
  30. 68.4 DeepSeek V4.1 Flash
  31. 66.5 Qwen3.8-2.4T-A95B
  32. 66.2 Ox Alpha
  33. 65.0 Gemini 3.1 Pro
  34. 64.0 Gemini 3.8 Flash
  35. 63.9 Glimpse
  36. 63.2 Gemini 3.5 Flash
  37. 63.1 Gemini 3.6 Flash
  38. 63.1 Ember-1
  39. 62.5 Claude Opus 4.8
  40. 62.2 Qwen3.8 Flash
  41. 62.1 GPT-5.4
  42. 61.5 Grok 4.5
  43. 60.8 GLM-5.3-Flash
  44. 60.8 Tencent Hy4 preview
  45. 59.9 GPT-5.5
  46. 59.3 Muse Spark Open
  47. 59.0 MiMo-V2.6 Flash
  48. 57.6 Mistral Large 4
  49. 57.3 Muse Spark 1.2
  50. 55.9 GPT-6 Luna
  51. 54.4 Claude Sonnet 5
  52. 54.2 Mistral Large 4 (Oct 29)
  53. 54.0 GPT-5.6 Luna
  54. 53.6 DeepSeek V4 Pro 0813
  55. 53.5 GPT-5.6 Terra
  56. 53.4 Beam 1.0 501B A23B
  57. 37.2 Nex N2 Pro
  58. 32.7 MiMo-V2.5
  59. 29.8 Ling-3.0-flash
  60. 21.5 Ripple-65B Preview
  61. 20.0 Nex N2.5 Pro
  62. 16.6 Qwen 3.8 27B
  63. 13.1 Solar Mini 4
  64. 11.2 Nemotron 3.5 Super VL
  65. 11.1 GPT-4.1 nano
  66. 5.40 North Mini Code 1.0
  67. 4.60 Granite 4.2 8B
  68. 0.00 Tencent Hy4
  1. 1 Claude Fable 5.2 ranked; XHigh 99.1
  2. 2 Claude Opus 5.5 ranked; XHigh 94.8
  3. 3 GPT-6 Astra ranked; XHigh 94.3
  4. 4 Claude Fable 5.1 ranked; XHigh 94.0
  5. 5 GPT-6.1 Sol ranked; XHigh 92.0
  6. 6 Claude Sonnet 5.5 ranked; XHigh 91.4
  7. 7 Pareto 26.10 ranked; Default 89.7
  8. 8 Gemini 4 Argon ranked; High 89.2
  9. 9 Pareto 26.10 Preview ranked; Default 88.3
  10. 10 GPT-6.1 Terra ranked; XHigh 85.8
  11. 11 GPT-6.1 Terra Pro ranked; XHigh 85.1
  12. 12 Claude Opus 5 ranked; XHigh 84.8
  13. 13 Kimi K3.1 Pro ranked; Max 84.0
  14. 14 GPT-5.6 Sol ranked; XHigh 80.1
  15. 15 GPT-6 Sol ranked; XHigh 79.4
  16. 16 Muse Wonder ranked; XHigh 77.8
  17. 17 Claude Fable 5 ranked; XHigh 75.2
  18. 18 Claude Haiku 5.5 ranked; XHigh 75.1
  19. 19 GLM-5.3 ranked; Max 72.3
  20. 20 Grok 4.7 ranked; Medium 72.3
  21. 21 Particle-1 ranked; Max 71.7
  22. 22 MiMo-V2.6 Pro ranked; Enabled 70.9
  23. 23 DeepSeek V4.1 Flash 1017 ranked; Max 69.9
  24. 24 GLM-5.3-FlashX-Legal ranked; Max 69.8
  25. 25 Qwen3.8-Omni-Flash pending; Enabled; Some games need a new attempt 69.7
  26. 26 Qwen3.8 Max (0902) ranked; XHigh 69.6
  27. 27 Grok 4.6 ranked; Medium 68.7
  28. 28 GLM-5.3 FlashX ranked; Max 68.5
  29. 29 Muse Spark 1.3 ranked; XHigh 68.5
  30. 30 DeepSeek V4.1 Flash ranked; Max 68.4
  31. 31 Qwen3.8-2.4T-A95B ranked; XHigh 66.5
  32. 32 Ox Alpha ranked; Max 66.2
  33. 33 Gemini 3.1 Pro ranked; High 65.0
  34. 34 Gemini 3.8 Flash ranked; High 64.0
  35. 35 Glimpse ranked; Enabled 63.9
  36. 36 Gemini 3.5 Flash ranked; High 63.2
  37. 37 Gemini 3.6 Flash ranked; High 63.1
  38. 38 Ember-1 ranked; Max 63.1
  39. 39 Claude Opus 4.8 ranked; XHigh 62.5
  40. 40 Qwen3.8 Flash ranked; Enabled 62.2
  41. 41 GPT-5.4 ranked; XHigh 62.1
  42. 42 Grok 4.5 ranked; High 61.5
  43. 43 GLM-5.3-Flash ranked; High 60.8
  44. 44 Tencent Hy4 preview ranked; High 60.8
  45. 45 GPT-5.5 ranked; XHigh 59.9
  46. 46 Muse Spark Open ranked; XHigh 59.3
  47. 47 MiMo-V2.6 Flash ranked; Enabled 59.0
  48. 48 Mistral Large 4 ranked; High 57.6
  49. 49 Muse Spark 1.2 ranked; Medium 57.3
  50. 50 GPT-6 Luna ranked; XHigh 55.9
  51. 51 Claude Sonnet 5 ranked; XHigh 54.4
  52. 52 Mistral Large 4 (Oct 29) ranked; Medium 54.2
  53. 53 GPT-5.6 Luna ranked; XHigh 54.0
  54. 54 DeepSeek V4 Pro 0813 ranked; Max 53.6
  55. 55 GPT-5.6 Terra ranked; XHigh 53.5
  56. 56 Beam 1.0 501B A23B ranked; XHigh 53.4
  57. 57 Nex N2 Pro pending; None; Can't be ranked: provider route no longer exists 37.2
  58. 58 MiMo-V2.5 pending; Enabled; Can't be ranked: original request cannot be reproduced 32.7
  59. 59 Ling-3.0-flash pending; Enabled; Can't be ranked: provider route no longer exists 29.8
  60. 60 Ripple-65B Preview pending; Default; Too many programs failed 21.5
  61. 61 Nex N2.5 Pro pending; High; Some games need a new attempt 20.0
  62. 62 Qwen 3.8 27B pending; None; Too many programs failed 16.6
  63. 63 Solar Mini 4 pending; XHigh; Too many programs failed 13.1
  64. 64 Nemotron 3.5 Super VL pending; None; Too many programs failed 11.2
  65. 65 GPT-4.1 nano pending; Omitted; Too many programs failed 11.1
  66. 66 North Mini Code 1.0 pending; None; Too many programs failed 5.40
  67. 67 Granite 4.2 8B pending; Low; Too many programs failed 4.60
  68. 68 Tencent Hy4 pending; High; Every attempt hit the 65,536-token output limit 0.00

Source: Official DuelLab GameBench 2 leaderboard

#EntryScoreSettingDateSource
1 Claude Fable 5.2 99.1 ranked; XHigh Nov 2026 Official leaderboard
2 Claude Opus 5.5 94.8 ranked; XHigh Sep 2026 Official leaderboard
3 GPT-6 Astra 94.3 ranked; XHigh Sep 2026 Official leaderboard
4 Claude Fable 5.1 94.0 ranked; XHigh Sep 2026 Official leaderboard
5 GPT-6.1 Sol 92.0 ranked; XHigh Sep 2026 Official leaderboard
6 Claude Sonnet 5.5 91.4 ranked; XHigh Sep 2026 Official leaderboard
7 Pareto 26.10 89.7 ranked; Default Official leaderboard
8 Gemini 4 Argon 89.2 ranked; High Sep 2026 Official leaderboard
9 Pareto 26.10 Preview 88.3 ranked; Default Official leaderboard
10 GPT-6.1 Terra 85.8 ranked; XHigh Oct 2026 Official leaderboard
11 GPT-6.1 Terra Pro 85.1 ranked; XHigh Official leaderboard
12 Claude Opus 5 84.8 ranked; XHigh Jul 2026 Official leaderboard
13 Kimi K3.1 Pro 84.0 ranked; Max Oct 2026 Official leaderboard
14 GPT-5.6 Sol 80.1 ranked; XHigh Jul 2026 Official leaderboard
15 GPT-6 Sol 79.4 ranked; XHigh Sep 2026 Official leaderboard
16 Muse Wonder 77.8 ranked; XHigh Oct 2026 Official leaderboard
17 Claude Fable 5 75.2 ranked; XHigh Jun 2026 Official leaderboard
18 Claude Haiku 5.5 75.1 ranked; XHigh Oct 2026 Official leaderboard
19 GLM-5.3 72.3 ranked; Max Aug 2026 Official leaderboard
20 Grok 4.7 72.3 ranked; Medium Sep 2026 Official leaderboard
21 Particle-1 71.7 ranked; Max Official leaderboard
22 MiMo-V2.6 Pro 70.9 ranked; Enabled Sep 2026 Official leaderboard
23 DeepSeek V4.1 Flash 1017 69.9 ranked; Max Oct 2026 Official leaderboard
24 GLM-5.3-FlashX-Legal 69.8 ranked; Max Oct 2026 Official leaderboard
25 Qwen3.8-Omni-Flash 69.7 pending; Enabled; Some games need a new attempt Sep 2026 Official leaderboard
26 Qwen3.8 Max (0902) 69.6 ranked; XHigh Sep 2026 Official leaderboard
27 Grok 4.6 68.7 ranked; Medium Aug 2026 Official leaderboard
28 GLM-5.3 FlashX 68.5 ranked; Max Official leaderboard
29 Muse Spark 1.3 68.5 ranked; XHigh Sep 2026 Official leaderboard
30 DeepSeek V4.1 Flash 68.4 ranked; Max Sep 2026 Official leaderboard
31 Qwen3.8-2.4T-A95B 66.5 ranked; XHigh Aug 2026 Official leaderboard
32 Ox Alpha 66.2 ranked; Max Official leaderboard
33 Gemini 3.1 Pro 65.0 ranked; High Feb 2026 Official leaderboard
34 Gemini 3.8 Flash 64.0 ranked; High Sep 2026 Official leaderboard
35 Glimpse 63.9 ranked; Enabled Official leaderboard
36 Gemini 3.5 Flash 63.2 ranked; High May 2026 Official leaderboard
37 Gemini 3.6 Flash 63.1 ranked; High Jul 2026 Official leaderboard
38 Ember-1 63.1 ranked; Max Sep 2026 Official leaderboard
39 Claude Opus 4.8 62.5 ranked; XHigh May 2026 Official leaderboard
40 Qwen3.8 Flash 62.2 ranked; Enabled Aug 2026 Official leaderboard
41 GPT-5.4 62.1 ranked; XHigh Mar 2026 Official leaderboard
42 Grok 4.5 61.5 ranked; High Jul 2026 Official leaderboard
43 GLM-5.3-Flash 60.8 ranked; High Aug 2026 Official leaderboard
44 Tencent Hy4 preview 60.8 ranked; High Aug 2026 Official leaderboard
45 GPT-5.5 59.9 ranked; XHigh Apr 2026 Official leaderboard
46 Muse Spark Open 59.3 ranked; XHigh Oct 2026 Official leaderboard
47 MiMo-V2.6 Flash 59.0 ranked; Enabled Sep 2026 Official leaderboard
48 Mistral Large 4 57.6 ranked; High Oct 2026 Official leaderboard
49 Muse Spark 1.2 57.3 ranked; Medium Aug 2026 Official leaderboard
50 GPT-6 Luna 55.9 ranked; XHigh Sep 2026 Official leaderboard
51 Claude Sonnet 5 54.4 ranked; XHigh Jun 2026 Official leaderboard
52 Mistral Large 4 (Oct 29) 54.2 ranked; Medium Official leaderboard
53 GPT-5.6 Luna 54.0 ranked; XHigh Jul 2026 Official leaderboard
54 DeepSeek V4 Pro 0813 53.6 ranked; Max Aug 2026 Official leaderboard
55 GPT-5.6 Terra 53.5 ranked; XHigh Jul 2026 Official leaderboard
56 Beam 1.0 501B A23B 53.4 ranked; XHigh Official leaderboard
57 Nex N2 Pro 37.2 pending; None; Can't be ranked: provider route no longer exists Official leaderboard
58 MiMo-V2.5 32.7 pending; Enabled; Can't be ranked: original request cannot be reproduced Apr 2026 Official leaderboard
59 Ling-3.0-flash 29.8 pending; Enabled; Can't be ranked: provider route no longer exists Aug 2026 Official leaderboard
60 Ripple-65B Preview 21.5 pending; Default; Too many programs failed Nov 2026 Official leaderboard
61 Nex N2.5 Pro 20.0 pending; High; Some games need a new attempt Official leaderboard
62 Qwen 3.8 27B 16.6 pending; None; Too many programs failed Aug 2026 Official leaderboard
63 Solar Mini 4 13.1 pending; XHigh; Too many programs failed Oct 2026 Official leaderboard
64 Nemotron 3.5 Super VL 11.2 pending; None; Too many programs failed Oct 2026 Official leaderboard
65 GPT-4.1 nano 11.1 pending; Omitted; Too many programs failed Apr 2025 Official leaderboard
66 North Mini Code 1.0 5.40 pending; None; Too many programs failed Jun 2026 Official leaderboard
67 Granite 4.2 8B 4.60 pending; Low; Too many programs failed Aug 2026 Official leaderboard
68 Tencent Hy4 0.00 pending; High; Every attempt hit the 65,536-token output limit Nov 2026 Official leaderboard

More effort adds a median 9 score

Models the leaderboard ran at two or more effort levels. We don’t have enough cost data here to say what the extra effort costs.

Lowest effort to highest, light to dark
Claude Fable 5.2
ranked; Low ranked; XHi…
99.1
Claude Opus 5.5
ranked; Low ranked; Med…
94.8
GPT-6 Astra
ranked; Low ranked; Med… ranked; XHi…
94.3
Claude Fable 5.1
ranked; Low ranked; XHi…
94.0
GPT-6.1 Sol
ranked; Low ranked; Med… ranked; XHi…
92.0
Gemini 4 Argon
ranked; Low ranked; High
89.2
GPT-6.1 Terra
ranked; None ranked; XHi…
85.8
Claude Opus 5
ranked; None ranked; Med… ranked; XHi…
84.8
Kimi K3.1 Pro
ranked; Low ranked; High ranked; Max
84.0
GPT-5.6 Sol
ranked; None ranked; Med… ranked; XHi…
80.1
GPT-6 Sol
ranked; None ranked; Med… ranked; XHi…
79.4
Muse Wonder
ranked; Min… ranked; Med… ranked; XHi…
77.8

Source: Official DuelLab GameBench 2 leaderboard

ModelEffortHarnessScoreCost per taskOutput tokens
Claude Fable 5.2 ranked; XHigh 99.1
Claude Opus 5.5 ranked; XHigh 94.8
GPT-6 Astra ranked; XHigh 94.3
Claude Fable 5.1 ranked; XHigh 94.0
Claude Fable 5.2 ranked; Medium 94.0
GPT-6.1 Sol ranked; XHigh 92.0
Claude Fable 5.2 ranked; Low 91.8
Claude Opus 5.5 ranked; Medium 91.4
Claude Sonnet 5.5 ranked; XHigh 91.4
Gemini 4 Argon ranked; High 89.2
GPT-6 Astra ranked; Medium 88.6
GPT-6.1 Sol ranked; Medium 86.4
GPT-6.1 Terra ranked; XHigh 85.8
Claude Opus 5 ranked; XHigh 84.8
Claude Fable 5.1 ranked; Low 84.7
Kimi K3.1 Pro ranked; Max 84.0
GPT-6 Astra ranked; Low 83.4
Claude Fable 5.1 ranked; Medium 82.9
GPT-5.6 Sol ranked; XHigh 80.1
GPT-6 Sol ranked; XHigh 79.4
Kimi K3.1 Pro ranked; Low 79.3
GPT-6.1 Terra ranked; None 79.3
Gemini 4 Argon ranked; Low 79.2
GPT-6.1 Terra ranked; Medium 78.3
Muse Wonder ranked; XHigh 77.8
Claude Opus 5.5 ranked; Low 76.6
Gemini 4 Argon ranked; Medium 76.6
GPT-6.1 Sol ranked; Low 76.4
Claude Fable 5 ranked; XHigh 75.2
Claude Haiku 5.5 ranked; XHigh 75.1
Claude Opus 5 ranked; Medium 75.0
GPT-6 Sol ranked; Medium 74.8
GLM-5.3 ranked; Max 72.3
Grok 4.7 ranked; Medium 72.3
Kimi K3.1 Pro ranked; High 71.5
Grok 4.7 ranked; XHigh 71.2
MiMo-V2.6 Pro ranked; Enabled 70.9
GLM-5.3 ranked; High 70.3
DeepSeek V4.1 Flash 1017 ranked; Max 69.9
Claude Opus 5 ranked; None 69.8
GLM-5.3-FlashX-Legal ranked; Max 69.8
Qwen3.8-Omni-Flash pending; Enabled; Some games need a new attempt 69.7
Qwen3.8 Max (0902) ranked; XHigh 69.6
Grok 4.6 ranked; Medium 68.7
Muse Spark 1.3 ranked; XHigh 68.5
DeepSeek V4.1 Flash ranked; Max 68.4
Qwen3.8-2.4T-A95B ranked; XHigh 66.5
GPT-5.6 Sol ranked; Medium 66.1
Grok 4.6 ranked; XHigh 65.6
Gemini 3.1 Pro ranked; High 65.0
GPT-6 Sol ranked; None 64.7
Muse Spark 1.3 ranked; Medium 64.5
DeepSeek V4.1 Flash 1017 ranked; High 64.3
Claude Fable 5 ranked; Medium 64.0
Gemini 3.8 Flash ranked; High 64.0
Gemini 3.5 Flash ranked; High 63.2
Gemini 3.6 Flash ranked; High 63.1
Ember-1 ranked; Max 63.1
Claude Opus 4.8 ranked; XHigh 62.5
Qwen3.8 Flash ranked; Enabled 62.2
GPT-5.4 ranked; XHigh 62.1
Muse Wonder ranked; Medium 61.6
Grok 4.5 ranked; High 61.5
GLM-5.3-Flash ranked; High 60.8
Tencent Hy4 preview ranked; High 60.8
Grok 4.5 ranked; Medium 60.3
GPT-5.5 ranked; XHigh 59.9
Gemini 3.8 Flash ranked; Medium 59.7
Muse Spark Open ranked; XHigh 59.3
MiMo-V2.6 Flash ranked; Enabled 59.0
GLM-5.3-FlashX-Legal ranked; High 58.9
Claude Fable 5 ranked; Low 58.2
Mistral Large 4 ranked; High 57.6
Muse Spark 1.2 ranked; Medium 57.3
GLM-5.3-Flash ranked; Max 57.2
Muse Spark 1.2 ranked; XHigh 56.9
Claude Haiku 5.5 ranked; Medium 56.2
GPT-6 Luna ranked; XHigh 55.9
Gemini 3.5 Flash ranked; Medium 55.4
Muse Wonder ranked; Minimal 54.9
GPT-5.6 Sol ranked; None 54.8
DeepSeek V4.1 Flash ranked; High 54.8
Claude Sonnet 5 ranked; XHigh 54.4
GPT-5.6 Luna ranked; XHigh 54.0
Muse Spark Open ranked; Low 54.0
DeepSeek V4 Pro 0813 ranked; Max 53.6
GPT-5.6 Terra ranked; XHigh 53.5
Gemini 3.8 Flash ranked; Low 53.5
GPT-5.4 ranked; Medium 53.3
MiMo-V2.5 pending; Enabled; Can't be ranked: original request cannot be reproduced 32.7
MiMo-V2.5 pending; High; Can't be ranked: original request cannot be reproduced 32.4
Ling-3.0-flash pending; Enabled; Can't be ranked: provider route no longer exists 29.8
MiMo-V2.5 pending; Medium; Can't be ranked: original request cannot be reproduced 28.5
Ripple-65B Preview pending; Default; Too many programs failed 21.5
Qwen 3.8 27B pending; None; Too many programs failed 16.6
Solar Mini 4 pending; XHigh; Too many programs failed 13.1
Nemotron 3.5 Super VL pending; None; Too many programs failed 11.2
GPT-4.1 nano pending; Omitted; Too many programs failed 11.1
North Mini Code 1.0 pending; None; Too many programs failed 5.40
Solar Mini 4 pending; Medium; Some games need a new attempt 5.30
Solar Mini 4 pending; None; Too many programs failed 5.00
Granite 4.2 8B pending; Low; Too many programs failed 4.60
Granite 4.2 8B pending; None; Every program failed 0.00
Granite 4.2 8B pending; Full; Every program failed 0.00
Tencent Hy4 pending; High; Every attempt hit the 65,536-token output limit 0.00
Tencent Hy4 pending; Low; Every attempt hit the 65,536-token output limit 0.00

7 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.

Anthropic OpenAI
Own leaderboard 7 leaders
Claude Fable 5.2 99.1

Source: Official DuelLab GameBench 2 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
0.00 20.0 40.0 60.0 80.0 100.0 2025 2026 2027 0.00 20.0 40.0 60.0 80.0 100.0 2025 2026 2027

Source: Official DuelLab GameBench 2 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%