Benchmark Almanac

BabaIsBench

GPT-5.5 100% leads since June 2026
Not in model cards

Agents play levels of the puzzle game Baba Is You through a text interface and win by rewriting the rules. Models are compared on pass rate, speed and cost.

23 levels in 2 stages and 32 models as of September 5th, 2026. Tasks never name the game. Quesma says OpenAI mentioned it with its ARC-AGI-3 results.

Made by
Quesma
Released
June 2026
Kind
Puzzles and chess
Level
Expert
Scored
Checked automatically
Answers
Agent task
Tasks
All public
Strongest model tested
GPT-6 Astra #3 on the capability index
Last model added
September 2026
Scores refreshed
November 12th, 2026
32
Models scored on its own leaderboard

8 models share the top score of 100% among models released by September 2026

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. Every run is in the next chart.

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

32 models. Scroll sideways for the rest.

  1. 100.0 GPT-5.5
  2. 100.0 Claude Fable 5
  3. 100.0 GPT-5.6 Sol
  4. 100.0 Claude Opus 5
  5. 100.0 Gemini 3.7 Flash
  6. 100.0 Claude Fable 5.1
  7. 100.0 Gemini 3.8 Flash
  8. 100.0 GPT-6 Astra
  9. 96.0 Claude Opus 4.8
  10. 96.0 GLM-5.2
  11. 96.0 Kimi K3
  12. 96.0 Grok 4.6
  13. 96.0 GLM-5.3
  14. 92.0 Gemini 3.1 Pro
  15. 88.0 Gemini 3.5 Flash
  16. 88.0 GPT-5.6 Terra
  17. 88.0 Gemini 3.6 Flash
  18. 75.0 DeepSeek-V4-Flash
  19. 75.0 Grok 4.5
  20. 75.0 DeepSeek V4 Flash 0731
  21. 75.0 DeepSeek V4 Pro 0813
  22. 67.0 Claude Sonnet 5
  23. 67.0 GLM-5.3-Flash
  24. 63.0 GPT-5.6 Luna
  25. 58.0 Qwen 3.8 Max
  26. 54.0 Tencent Hy3
  27. 46.0 MiniMax-M3
  28. 46.0 Muse Spark 1.3
  29. 29.0 Qwen3.6 27B
  30. 29.0 DeepSeek-V4-Pro
  31. 25.0 Qwen3.7-Max
  32. 21.0 Qwen 3.8 27B
  1. 1 GPT-5.5 Stage 0; 8 levels; 3 attempts 100%
  2. 2 Claude Fable 5 Stage 0; 8 levels; 3 attempts 100%
  3. 3 GPT-5.6 Sol Stage 0; 8 levels; 3 attempts 100%
  4. 4 Claude Opus 5 Stage 0; 8 levels; 3 attempts 100%
  5. 5 Gemini 3.7 Flash Stage 0; 8 levels; 3 attempts 100%
  6. 6 Claude Fable 5.1 Stage 0; 8 levels; 3 attempts 100%
  7. 7 Gemini 3.8 Flash Stage 0; 8 levels; 3 attempts 100%
  8. 8 GPT-6 Astra Stage 0; 8 levels; 3 attempts 100%
  9. 9 Claude Opus 4.8 Stage 0; 8 levels; 3 attempts 96%
  10. 10 GLM-5.2 Stage 0; 8 levels; 3 attempts 96%
  11. 11 Kimi K3 Stage 0; 8 levels; 3 attempts 96%
  12. 12 Grok 4.6 Stage 0; 8 levels; 3 attempts 96%
  13. 13 GLM-5.3 Stage 0; 8 levels; 3 attempts 96%
  14. 14 Gemini 3.1 Pro Stage 0; 8 levels; 3 attempts 92%
  15. 15 Gemini 3.5 Flash Stage 0; 8 levels; 3 attempts 88%
  16. 16 GPT-5.6 Terra Stage 0; 8 levels; 3 attempts 88%
  17. 17 Gemini 3.6 Flash Stage 0; 8 levels; 3 attempts 88%
  18. 18 DeepSeek-V4-Flash Stage 0; 8 levels; 3 attempts 75%
  19. 19 Grok 4.5 Stage 0; 8 levels; 3 attempts 75%
  20. 20 DeepSeek V4 Flash 0731 Stage 0; 8 levels; 3 attempts 75%
  21. 21 DeepSeek V4 Pro 0813 Stage 0; 8 levels; 3 attempts 75%
  22. 22 Claude Sonnet 5 Stage 0; 8 levels; 3 attempts 67%
  23. 23 GLM-5.3-Flash Stage 0; 8 levels; 3 attempts 67%
  24. 24 GPT-5.6 Luna Stage 0; 8 levels; 3 attempts 63%
  25. 25 Qwen 3.8 Max Stage 0; 8 levels; 3 attempts 58%
  26. 26 Tencent Hy3 Stage 0; 8 levels; 3 attempts 54%
  27. 27 MiniMax-M3 Stage 0; 8 levels; 3 attempts 46%
  28. 28 Muse Spark 1.3 Stage 0; 8 levels; 3 attempts 46%
  29. 29 Qwen3.6 27B Stage 0; 8 levels; 3 attempts 29%
  30. 30 DeepSeek-V4-Pro Stage 0; 8 levels; 3 attempts 29%
  31. 31 Qwen3.7-Max Stage 0; 8 levels; 3 attempts 25%
  32. 32 Qwen 3.8 27B Stage 0; 8 levels; 3 attempts 21%

Source: Official BabaIsBench leaderboard

#ModelScoreSettingDateSource
1 GPT-5.5 100% Stage 0; 8 levels; 3 attempts Apr 2026 Official leaderboard
2 Claude Fable 5 100% Stage 0; 8 levels; 3 attempts Jun 2026 Official leaderboard
3 GPT-5.6 Sol 100% Stage 0; 8 levels; 3 attempts Jul 2026 Official leaderboard
4 Claude Opus 5 100% Stage 0; 8 levels; 3 attempts Jul 2026 Official leaderboard
5 Gemini 3.7 Flash 100% Stage 0; 8 levels; 3 attempts Aug 2026 Official leaderboard
6 Claude Fable 5.1 100% Stage 0; 8 levels; 3 attempts Sep 2026 Official leaderboard
7 Gemini 3.8 Flash 100% Stage 0; 8 levels; 3 attempts Sep 2026 Official leaderboard
8 GPT-6 Astra 100% Stage 0; 8 levels; 3 attempts Sep 2026 Official leaderboard
9 Claude Opus 4.8 96% Stage 0; 8 levels; 3 attempts May 2026 Official leaderboard
10 GLM-5.2 96% Stage 0; 8 levels; 3 attempts Jun 2026 Official leaderboard
11 Kimi K3 96% Stage 0; 8 levels; 3 attempts Jul 2026 Official leaderboard
12 Grok 4.6 96% Stage 0; 8 levels; 3 attempts Aug 2026 Official leaderboard
13 GLM-5.3 96% Stage 0; 8 levels; 3 attempts Aug 2026 Official leaderboard
14 Gemini 3.1 Pro 92% Stage 0; 8 levels; 3 attempts Feb 2026 Official leaderboard
15 Gemini 3.5 Flash 88% Stage 0; 8 levels; 3 attempts May 2026 Official leaderboard
16 GPT-5.6 Terra 88% Stage 0; 8 levels; 3 attempts Jul 2026 Official leaderboard
17 Gemini 3.6 Flash 88% Stage 0; 8 levels; 3 attempts Jul 2026 Official leaderboard
18 DeepSeek-V4-Flash 75% Stage 0; 8 levels; 3 attempts Apr 2026 Official leaderboard
19 Grok 4.5 75% Stage 0; 8 levels; 3 attempts Jul 2026 Official leaderboard
20 DeepSeek V4 Flash 0731 75% Stage 0; 8 levels; 3 attempts Jul 2026 Official leaderboard
21 DeepSeek V4 Pro 0813 75% Stage 0; 8 levels; 3 attempts Aug 2026 Official leaderboard
22 Claude Sonnet 5 67% Stage 0; 8 levels; 3 attempts Jun 2026 Official leaderboard
23 GLM-5.3-Flash 67% Stage 0; 8 levels; 3 attempts Aug 2026 Official leaderboard
24 GPT-5.6 Luna 63% Stage 0; 8 levels; 3 attempts Jul 2026 Official leaderboard
25 Qwen 3.8 Max 58% Stage 0; 8 levels; 3 attempts Aug 2026 Official leaderboard
26 Tencent Hy3 54% Stage 0; 8 levels; 3 attempts Jul 2026 Official leaderboard
27 MiniMax-M3 46% Stage 0; 8 levels; 3 attempts Jun 2026 Official leaderboard
28 Muse Spark 1.3 46% Stage 0; 8 levels; 3 attempts Sep 2026 Official leaderboard
29 Qwen3.6 27B 29% Stage 0; 8 levels; 3 attempts Apr 2026 Official leaderboard
30 DeepSeek-V4-Pro 29% Stage 0; 8 levels; 3 attempts Apr 2026 Official leaderboard
31 Qwen3.7-Max 25% Stage 0; 8 levels; 3 attempts May 2026 Official leaderboard
32 Qwen 3.8 27B 21% Stage 0; 8 levels; 3 attempts Aug 2026 Official leaderboard

Intro levels cost frontier

The line joins the runs that no other run beats on both measures. We redrew this chart from the numbers on the official leaderboard.

0% 20% 40% 60% 80% 100% $0.1$1$10$100$1,000 Total cost of 24 attempts ($), log scale, lower is better pass@1 Gemini 3.1 Pro (Stage 0; 8 levels; 3 attempts): pass@1 92%, total cost of 24 attempts $12.4 Qwen3.6 27B (Stage 0; 8 levels; 3 attempts): pass@1 29%, total cost of 24 attempts $5.59 GPT-5.5 (Stage 0; 8 levels; 3 attempts): pass@1 100%, total cost of 24 attempts $13.6 DeepSeek-V4-Flash (Stage 0; 8 levels; 3 attempts): pass@1 75%, total cost of 24 attempts $2.82 DeepSeek-V4-Pro (Stage 0; 8 levels; 3 attempts): pass@1 29%, total cost of 24 attempts $11.5 Gemini 3.5 Flash (Stage 0; 8 levels; 3 attempts): pass@1 88%, total cost of 24 attempts $98.3 Qwen3.7-Max (Stage 0; 8 levels; 3 attempts): pass@1 25%, total cost of 24 attempts $17 Claude Opus 4.8 (Stage 0; 8 levels; 3 attempts): pass@1 96%, total cost of 24 attempts $63.1 MiniMax-M3 (Stage 0; 8 levels; 3 attempts): pass@1 46%, total cost of 24 attempts $34.4 Claude Fable 5 (Stage 0; 8 levels; 3 attempts): pass@1 100%, total cost of 24 attempts $41 GLM-5.2 (Stage 0; 8 levels; 3 attempts): pass@1 96%, total cost of 24 attempts $6.2 Claude Sonnet 5 (Stage 0; 8 levels; 3 attempts): pass@1 67%, total cost of 24 attempts $42.2 Tencent Hy3 (Stage 0; 8 levels; 3 attempts): pass@1 54%, total cost of 24 attempts $3.23 Grok 4.5 (Stage 0; 8 levels; 3 attempts): pass@1 75%, total cost of 24 attempts $50.7 GPT-5.6 Luna (Stage 0; 8 levels; 3 attempts): pass@1 63%, total cost of 24 attempts $59.3 GPT-5.6 Sol (Stage 0; 8 levels; 3 attempts): pass@1 100%, total cost of 24 attempts $11.9 GPT-5.6 Terra (Stage 0; 8 levels; 3 attempts): pass@1 88%, total cost of 24 attempts $34.3 Kimi K3 (Stage 0; 8 levels; 3 attempts): pass@1 96%, total cost of 24 attempts $12.6 Gemini 3.6 Flash (Stage 0; 8 levels; 3 attempts): pass@1 88%, total cost of 24 attempts $124 Claude Opus 5 (Stage 0; 8 levels; 3 attempts): pass@1 100%, total cost of 24 attempts $17.1 Qwen 3.8 Max (Stage 0; 8 levels; 3 attempts): pass@1 58%, total cost of 24 attempts $10.8 Grok 4.6 (Stage 0; 8 levels; 3 attempts): pass@1 96%, total cost of 24 attempts $14 DeepSeek V4 Pro 0813 (Stage 0; 8 levels; 3 attempts): pass@1 75%, total cost of 24 attempts $2.94 Gemini 3.7 Flash (Stage 0; 8 levels; 3 attempts): pass@1 100%, total cost of 24 attempts $5.38 GLM-5.3 (Stage 0; 8 levels; 3 attempts): pass@1 96%, total cost of 24 attempts $6.2 Qwen 3.8 27B (Stage 0; 8 levels; 3 attempts): pass@1 21%, total cost of 24 attempts $9.07 Claude Fable 5.1 (Stage 0; 8 levels; 3 attempts): pass@1 100%, total cost of 24 attempts $13.2 Gemini 3.8 Flash (Stage 0; 8 levels; 3 attempts): pass@1 100%, total cost of 24 attempts $5.93 Muse Spark 1.3 (Stage 0; 8 levels; 3 attempts): pass@1 46%, total cost of 24 attempts $30.9 DeepSeek V4 Flash 0731 (Stage 0; 8 levels; 3 attempts): pass@1 75%, total cost of 24 attempts $1.16 GLM-5.3-Flash (Stage 0; 8 levels; 3 attempts): pass@1 67%, total cost of 24 attempts $0.32 GPT-6 Astra (Stage 0; 8 levels; 3 attempts): pass@1 100%, total cost of 24 attempts $5.33 GPT-6 Astra (Stage 0; 8 levels; 3 attempts)DeepSeek V4 Flash 0731 (Stage 0; 8 levels; 3 attempts)GLM-5.3-Flash (Stage 0; 8 levels; 3 attempts) 0% 20% 40% 60% 80% 100% $0.1$1$10$100$1,000 Total cost of 24 attempts ($), log scale, lower is better pass@1 Gemini 3.1 Pro (Stage 0; 8 levels; 3 attempts): pass@1 92%, total cost of 24 attempts $12.4 Qwen3.6 27B (Stage 0; 8 levels; 3 attempts): pass@1 29%, total cost of 24 attempts $5.59 GPT-5.5 (Stage 0; 8 levels; 3 attempts): pass@1 100%, total cost of 24 attempts $13.6 DeepSeek-V4-Flash (Stage 0; 8 levels; 3 attempts): pass@1 75%, total cost of 24 attempts $2.82 DeepSeek-V4-Pro (Stage 0; 8 levels; 3 attempts): pass@1 29%, total cost of 24 attempts $11.5 Gemini 3.5 Flash (Stage 0; 8 levels; 3 attempts): pass@1 88%, total cost of 24 attempts $98.3 Qwen3.7-Max (Stage 0; 8 levels; 3 attempts): pass@1 25%, total cost of 24 attempts $17 Claude Opus 4.8 (Stage 0; 8 levels; 3 attempts): pass@1 96%, total cost of 24 attempts $63.1 MiniMax-M3 (Stage 0; 8 levels; 3 attempts): pass@1 46%, total cost of 24 attempts $34.4 Claude Fable 5 (Stage 0; 8 levels; 3 attempts): pass@1 100%, total cost of 24 attempts $41 GLM-5.2 (Stage 0; 8 levels; 3 attempts): pass@1 96%, total cost of 24 attempts $6.2 Claude Sonnet 5 (Stage 0; 8 levels; 3 attempts): pass@1 67%, total cost of 24 attempts $42.2 Tencent Hy3 (Stage 0; 8 levels; 3 attempts): pass@1 54%, total cost of 24 attempts $3.23 Grok 4.5 (Stage 0; 8 levels; 3 attempts): pass@1 75%, total cost of 24 attempts $50.7 GPT-5.6 Luna (Stage 0; 8 levels; 3 attempts): pass@1 63%, total cost of 24 attempts $59.3 GPT-5.6 Sol (Stage 0; 8 levels; 3 attempts): pass@1 100%, total cost of 24 attempts $11.9 GPT-5.6 Terra (Stage 0; 8 levels; 3 attempts): pass@1 88%, total cost of 24 attempts $34.3 Kimi K3 (Stage 0; 8 levels; 3 attempts): pass@1 96%, total cost of 24 attempts $12.6 Gemini 3.6 Flash (Stage 0; 8 levels; 3 attempts): pass@1 88%, total cost of 24 attempts $124 Claude Opus 5 (Stage 0; 8 levels; 3 attempts): pass@1 100%, total cost of 24 attempts $17.1 Qwen 3.8 Max (Stage 0; 8 levels; 3 attempts): pass@1 58%, total cost of 24 attempts $10.8 Grok 4.6 (Stage 0; 8 levels; 3 attempts): pass@1 96%, total cost of 24 attempts $14 DeepSeek V4 Pro 0813 (Stage 0; 8 levels; 3 attempts): pass@1 75%, total cost of 24 attempts $2.94 Gemini 3.7 Flash (Stage 0; 8 levels; 3 attempts): pass@1 100%, total cost of 24 attempts $5.38 GLM-5.3 (Stage 0; 8 levels; 3 attempts): pass@1 96%, total cost of 24 attempts $6.2 Qwen 3.8 27B (Stage 0; 8 levels; 3 attempts): pass@1 21%, total cost of 24 attempts $9.07 Claude Fable 5.1 (Stage 0; 8 levels; 3 attempts): pass@1 100%, total cost of 24 attempts $13.2 Gemini 3.8 Flash (Stage 0; 8 levels; 3 attempts): pass@1 100%, total cost of 24 attempts $5.93 Muse Spark 1.3 (Stage 0; 8 levels; 3 attempts): pass@1 46%, total cost of 24 attempts $30.9 DeepSeek V4 Flash 0731 (Stage 0; 8 levels; 3 attempts): pass@1 75%, total cost of 24 attempts $1.16 GLM-5.3-Flash (Stage 0; 8 levels; 3 attempts): pass@1 67%, total cost of 24 attempts $0.32 GPT-6 Astra (Stage 0; 8 levels; 3 attempts): pass@1 100%, total cost of 24 attempts $5.33 GPT-6 Astra (Stage 0; 8 levels; 3 attempts)DeepSeek V4 Flash 0731 (Stage 0; 8 levels; 3 attempts)GLM-5.3-Flash (Stage 0; 8 levels; 3 attempts)
ModelSettingpass@1Total cost of 24 attempts
GPT-6 Astra Stage 0; 8 levels; 3 attempts 100% $5.33
Gemini 3.7 Flash Stage 0; 8 levels; 3 attempts 100% $5.38
Gemini 3.8 Flash Stage 0; 8 levels; 3 attempts 100% $5.93
GPT-5.6 Sol Stage 0; 8 levels; 3 attempts 100% $11.91
Claude Fable 5.1 Stage 0; 8 levels; 3 attempts 100% $13.18
GPT-5.5 Stage 0; 8 levels; 3 attempts 100% $13.62
Claude Opus 5 Stage 0; 8 levels; 3 attempts 100% $17.15
Claude Fable 5 Stage 0; 8 levels; 3 attempts 100% $40.98
GLM-5.2 Stage 0; 8 levels; 3 attempts 96% $6.2
GLM-5.3 Stage 0; 8 levels; 3 attempts 96% $6.2
Kimi K3 Stage 0; 8 levels; 3 attempts 96% $12.56
Grok 4.6 Stage 0; 8 levels; 3 attempts 96% $14.01
Claude Opus 4.8 Stage 0; 8 levels; 3 attempts 96% $63.14
Gemini 3.1 Pro Stage 0; 8 levels; 3 attempts 92% $12.41
GPT-5.6 Terra Stage 0; 8 levels; 3 attempts 88% $34.28
Gemini 3.5 Flash Stage 0; 8 levels; 3 attempts 88% $98.31
Gemini 3.6 Flash Stage 0; 8 levels; 3 attempts 88% $124.31
DeepSeek V4 Flash 0731 Stage 0; 8 levels; 3 attempts 75% $1.16
DeepSeek-V4-Flash Stage 0; 8 levels; 3 attempts 75% $2.82
DeepSeek V4 Pro 0813 Stage 0; 8 levels; 3 attempts 75% $2.94
Grok 4.5 Stage 0; 8 levels; 3 attempts 75% $50.66
GLM-5.3-Flash Stage 0; 8 levels; 3 attempts 67% $0.32
Claude Sonnet 5 Stage 0; 8 levels; 3 attempts 67% $42.21
GPT-5.6 Luna Stage 0; 8 levels; 3 attempts 63% $59.32
Qwen 3.8 Max Stage 0; 8 levels; 3 attempts 58% $10.75
Tencent Hy3 Stage 0; 8 levels; 3 attempts 54% $3.23
Muse Spark 1.3 Stage 0; 8 levels; 3 attempts 46% $30.9
MiniMax-M3 Stage 0; 8 levels; 3 attempts 46% $34.43
Qwen3.6 27B Stage 0; 8 levels; 3 attempts 29% $5.59
DeepSeek-V4-Pro Stage 0; 8 levels; 3 attempts 29% $11.49
Qwen3.7-Max Stage 0; 8 levels; 3 attempts 25% $16.98
Qwen 3.8 27B Stage 0; 8 levels; 3 attempts 21% $9.07

The best score, 100%, costs $0.22 a task

Runs on the cost frontier score higher than any run that's cheaper. The costs are the leaderboard's own estimates, so compare them only within this benchmark.

DeepSeek Z.ai OpenAI Cost frontier One model at each effort level
0% 20% 40% 60% 80% 100% $0.01 $0.10 $1.00 $10 Cost per task, log scale GPT-6 Astra, Stage 0; 8 levels; 3 attempts DeepSeek V4 Flash 0731, Stage 0; 8 levels; 3 attempts GLM-5.3-Flash, Stage 0; 8 levels; 3 attempts 0% 20% 40% 60% 80% 100% $0.01 $0.10 $1.00 $10 Cost per task, log scale GPT-6 Astra, Stage 0; 8 levels; 3 attempts DeepSeek V4 Flash 0731, Stage 0; 8 levels; 3 attempts

Source: Official BabaIsBench leaderboard

ModelEffortHarnessScoreCost per taskOutput tokens
GPT-5.5 Stage 0; 8 levels; 3 attempts 100% $0.57
Claude Fable 5 Stage 0; 8 levels; 3 attempts 100% $1.71
GPT-5.6 Sol Stage 0; 8 levels; 3 attempts 100% $0.50
Claude Opus 5 Stage 0; 8 levels; 3 attempts 100% $0.71
Gemini 3.7 Flash Stage 0; 8 levels; 3 attempts 100% $0.22
Claude Fable 5.1 Stage 0; 8 levels; 3 attempts 100% $0.55
Gemini 3.8 Flash Stage 0; 8 levels; 3 attempts 100% $0.25
GPT-6 Astra Stage 0; 8 levels; 3 attempts 100% $0.22
Claude Opus 4.8 Stage 0; 8 levels; 3 attempts 96% $2.63
GLM-5.2 Stage 0; 8 levels; 3 attempts 96% $0.26
Kimi K3 Stage 0; 8 levels; 3 attempts 96% $0.52
Grok 4.6 Stage 0; 8 levels; 3 attempts 96% $0.58
GLM-5.3 Stage 0; 8 levels; 3 attempts 96% $0.26
Gemini 3.1 Pro Stage 0; 8 levels; 3 attempts 92% $0.52
Gemini 3.5 Flash Stage 0; 8 levels; 3 attempts 88% $4.10
GPT-5.6 Terra Stage 0; 8 levels; 3 attempts 88% $1.43
Gemini 3.6 Flash Stage 0; 8 levels; 3 attempts 88% $5.18
DeepSeek-V4-Flash Stage 0; 8 levels; 3 attempts 75% $0.12
Grok 4.5 Stage 0; 8 levels; 3 attempts 75% $2.11
DeepSeek V4 Flash 0731 Stage 0; 8 levels; 3 attempts 75% $0.05
DeepSeek V4 Pro 0813 Stage 0; 8 levels; 3 attempts 75% $0.12
Claude Sonnet 5 Stage 0; 8 levels; 3 attempts 67% $1.76
GLM-5.3-Flash Stage 0; 8 levels; 3 attempts 67% $0.01
GPT-5.6 Luna Stage 0; 8 levels; 3 attempts 63% $2.47
Qwen 3.8 Max Stage 0; 8 levels; 3 attempts 58% $0.45
Tencent Hy3 Stage 0; 8 levels; 3 attempts 54% $0.13
MiniMax-M3 Stage 0; 8 levels; 3 attempts 46% $1.44
Muse Spark 1.3 Stage 0; 8 levels; 3 attempts 46% $1.29
Qwen3.6 27B Stage 0; 8 levels; 3 attempts 29% $0.23
DeepSeek-V4-Pro Stage 0; 8 levels; 3 attempts 29% $0.48
Qwen3.7-Max Stage 0; 8 levels; 3 attempts 25% $0.71
Qwen 3.8 27B Stage 0; 8 levels; 3 attempts 21% $0.38

Number one over time

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

OpenAI
Own leaderboard Passed 85% in Jun 2026, by GPT-5.5
GPT-5.5 100%

Source: Official BabaIsBench leaderboard

Gemini 3.1 Pro on the official leaderboard passed 85% in February 2026

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 Best so far
0% 20% 40% 60% 80% 100% 2026 Near the ceiling (85%) GPT-5.5 Gemini 3.1 Pro 0% 20% 40% 60% 80% 100% 2026 Near the ceiling (85%) GPT-5.5 Gemini 3.1 Pro

Source: Official BabaIsBench 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%