Benchmark Almanac

MineBench

GPT-6 Astra Pro 2,206 leads since September 2026
Not in model cards

Models turn a prompt into a Minecraft-style voxel build as raw 3D coordinates or tool calls. Users compare builds blind, and votes become rankings.

The maker says it is an arena of vibes with no correct answers. Kaggle also hosts a MineBench Arena board. About 340 GitHub stars.

Made by
Ammaar Alam
Released
January 2026
Kind
Spatial reasoning and counting
Level
Everyday
Scored
Ranked by votes
Answers
Open-ended
Tasks
All public
Strongest model tested
Claude Fable 5.2 #1 on the capability index
Last model added
November 2026
Scores refreshed
November 13th, 2026
80
Entries on its own leaderboard, 75 models we track

GPT-6 Astra Pro leads with 2,206

Best score per entry from the benchmark's own leaderboard, read on November 13th, 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.

80 entries. Scroll sideways for the rest. Bars start at 500.

  1. 2,206 GPT-6 Astra Pro
  2. 2,195 Gemini 4 Carbon
  3. 2,172 Gemini 4 Argon
  4. 2,147 GPT 6.1 Sol Pro
  5. 2,112 Claude Opus 5.5
  6. 2,102 Claude Fable 5.2
  7. 2,046 GPT 6 Sol Pro
  8. 1,993 Gemini 4 Hydrogen
  9. 1,988 Claude Opus 5
  10. 1,977 Claude Sonnet 5.5
  11. 1,973 GPT-5.6 Sol Pro
  12. 1,970 Kimi K3.1 Pro
  13. 1,909 Claude Fable 5.1
  14. 1,903 GPT-5.5 Pro
  15. 1,851 Claude Fable 5
  16. 1,850 Muse Wonder
  17. 1,829 Grok 4.7
  18. 1,828 GPT-5.5
  19. 1,804 DeepSeek V4.1 Flash
  20. 1,793 Grok 4.6
  21. 1,789 Gemini 3.8 Flash
  22. 1,777 GPT-5.4 Pro
  23. 1,768 Claude Opus 4.8
  24. 1,763 GLM-5.3-Flash
  25. 1,748 Gemini 3.7 Flash
  26. 1,740 GLM-5.3
  27. 1,730 Muse Spark 1.3
  28. 1,728 GPT 6 Luna Pro
  29. 1,697 Kimi K3
  30. 1,686 GPT 5.6 Luna Pro
  31. 1,661 Gemini 3.1 Pro
  32. 1,651 GPT-5.4
  33. 1,636 Gemini 3.5 Flash
  34. 1,610 Claude Sonnet 5
  35. 1,603 Gemini 3.6 Flash
  36. 1,593 Claude Opus 4.7
  37. 1,565 Claude Haiku 5.5
  38. 1,564 GPT-5.3 Codex
  39. 1,552 Muse Spark 1.2
  40. 1,547 GLM-5.2
  41. 1,539 GPT-5.2 Pro
  42. 1,538 Grok 4.5
  43. 1,501 GLM-5.1
  44. 1,482 DeepSeek V4 Flash 0731
  45. 1,472 Claude Opus 4.6
  46. 1,456 GPT-5.2
  47. 1,442 DeepSeek-V4-Pro
  48. 1,425 Kimi K2.6
  49. 1,387 GPT-5.4 Mini
  50. 1,375 Gemini 3 Pro
  51. 1,345 Claude Sonnet 4.6
  52. 1,339 Gemini 3.5 Flash-Lite
  53. 1,308 Claude Opus 4.5
  54. 1,280 Grok 4.20
  55. 1,275 Qwen3.5 397B-A17B
  56. 1,248 GPT-5.4 Nano
  57. 1,244 Gemini 3 Flash
  58. 1,218 Qwen 3.8 Max
  59. 1,214 GLM-5
  60. 1,213 Gemma 4 31B IT
  61. 1,194 MiniMax-M2.7
  62. 1,187 Kimi K2.5
  63. 1,147 Grok 4.3
  64. 1,140 Claude Sonnet 4.5
  65. 1,128 Gemini 2.5 Pro
  66. 1,100 GLM-4.7
  67. 1,076 GPT-5 mini
  68. 1,068 MiniMax-M2.5
  69. 1,052 GPT-5.2 Codex
  70. 1,037 DeepSeek-V3.2
  71. 1,033 GPT-4.1
  72. 990 Qwen3-Max
  73. 963 Kimi K2
  74. 953 Grok 4.1 Fast
  75. 944 Gemini 3.1 Flash-Lite
  76. 917 GPT 4.5 (web harness)
  77. 777 gpt-oss-120b
  78. 744 GPT-4o
  79. 628 Llama 4 Maverick
  80. 563 GPT-5 nano

Bars start at 500.

  1. 1 GPT-6 Astra Pro 2,206
  2. 2 Gemini 4 Carbon 2,195
  3. 3 Gemini 4 Argon 2,172
  4. 4 GPT 6.1 Sol Pro 2,147
  5. 5 Claude Opus 5.5 2,112
  6. 6 Claude Fable 5.2 2,102
  7. 7 GPT 6 Sol Pro 2,046
  8. 8 Gemini 4 Hydrogen 1,993
  9. 9 Claude Opus 5 1,988
  10. 10 Claude Sonnet 5.5 1,977
  11. 11 GPT-5.6 Sol Pro 1,973
  12. 12 Kimi K3.1 Pro 1,970
  13. 13 Claude Fable 5.1 1,909
  14. 14 GPT-5.5 Pro 1,903
  15. 15 Claude Fable 5 1,851
  16. 16 Muse Wonder 1,850
  17. 17 Grok 4.7 1,829
  18. 18 GPT-5.5 1,828
  19. 19 DeepSeek V4.1 Flash 1,804
  20. 20 Grok 4.6 1,793
  21. 21 Gemini 3.8 Flash 1,789
  22. 22 GPT-5.4 Pro 1,777
  23. 23 Claude Opus 4.8 1,768
  24. 24 GLM-5.3-Flash 1,763
  25. 25 Gemini 3.7 Flash 1,748
  26. 26 GLM-5.3 1,740
  27. 27 Muse Spark 1.3 1,730
  28. 28 GPT 6 Luna Pro 1,728
  29. 29 Kimi K3 1,697
  30. 30 GPT 5.6 Luna Pro 1,686
  31. 31 Gemini 3.1 Pro 1,661
  32. 32 GPT-5.4 1,651
  33. 33 Gemini 3.5 Flash 1,636
  34. 34 Claude Sonnet 5 1,610
  35. 35 Gemini 3.6 Flash 1,603
  36. 36 Claude Opus 4.7 1,593
  37. 37 Claude Haiku 5.5 1,565
  38. 38 GPT-5.3 Codex 1,564
  39. 39 Muse Spark 1.2 1,552
  40. 40 GLM-5.2 1,547
  41. 41 GPT-5.2 Pro 1,539
  42. 42 Grok 4.5 1,538
  43. 43 GLM-5.1 1,501
  44. 44 DeepSeek V4 Flash 0731 1,482
  45. 45 Claude Opus 4.6 1,472
  46. 46 GPT-5.2 1,456
  47. 47 DeepSeek-V4-Pro 1,442
  48. 48 Kimi K2.6 1,425
  49. 49 GPT-5.4 Mini 1,387
  50. 50 Gemini 3 Pro 1,375
  51. 51 Claude Sonnet 4.6 1,345
  52. 52 Gemini 3.5 Flash-Lite 1,339
  53. 53 Claude Opus 4.5 1,308
  54. 54 Grok 4.20 1,280
  55. 55 Qwen3.5 397B-A17B 1,275
  56. 56 GPT-5.4 Nano 1,248
  57. 57 Gemini 3 Flash 1,244
  58. 58 Qwen 3.8 Max 1,218
  59. 59 GLM-5 1,214
  60. 60 Gemma 4 31B IT 1,213
  61. 61 MiniMax-M2.7 1,194
  62. 62 Kimi K2.5 1,187
  63. 63 Grok 4.3 1,147
  64. 64 Claude Sonnet 4.5 1,140
  65. 65 Gemini 2.5 Pro Jun 2025 1,128
  66. 66 GLM-4.7 1,100
  67. 67 GPT-5 mini 1,076
  68. 68 MiniMax-M2.5 1,068
  69. 69 GPT-5.2 Codex 1,052
  70. 70 DeepSeek-V3.2 1,037
  71. 71 GPT-4.1 1,033
  72. 72 Qwen3-Max 990
  73. 73 Kimi K2 Sep 2025 963
  74. 74 Grok 4.1 Fast 953
  75. 75 Gemini 3.1 Flash-Lite 944
  76. 76 GPT 4.5 (web harness) 917
  77. 77 gpt-oss-120b 777
  78. 78 GPT-4o Mar 2025 744
  79. 79 Llama 4 Maverick 628
  80. 80 GPT-5 nano 563

Source: Official MineBench leaderboard

#EntryScoreSettingDateSource
1 GPT-6 Astra Pro 2,206 Sep 2026 Official leaderboard
2 Gemini 4 Carbon 2,195 Oct 2026 Official leaderboard
3 Gemini 4 Argon 2,172 Sep 2026 Official leaderboard
4 GPT 6.1 Sol Pro 2,147 Official leaderboard
5 Claude Opus 5.5 2,112 Sep 2026 Official leaderboard
6 Claude Fable 5.2 2,102 Nov 2026 Official leaderboard
7 GPT 6 Sol Pro 2,046 Official leaderboard
8 Gemini 4 Hydrogen 1,993 Nov 2026 Official leaderboard
9 Claude Opus 5 1,988 Jul 2026 Official leaderboard
10 Claude Sonnet 5.5 1,977 Sep 2026 Official leaderboard
11 GPT-5.6 Sol Pro 1,973 Jul 2026 Official leaderboard
12 Kimi K3.1 Pro 1,970 Oct 2026 Official leaderboard
13 Claude Fable 5.1 1,909 Sep 2026 Official leaderboard
14 GPT-5.5 Pro 1,903 Apr 2026 Official leaderboard
15 Claude Fable 5 1,851 Jun 2026 Official leaderboard
16 Muse Wonder 1,850 Oct 2026 Official leaderboard
17 Grok 4.7 1,829 Sep 2026 Official leaderboard
18 GPT-5.5 1,828 Apr 2026 Official leaderboard
19 DeepSeek V4.1 Flash 1,804 Sep 2026 Official leaderboard
20 Grok 4.6 1,793 Aug 2026 Official leaderboard
21 Gemini 3.8 Flash 1,789 Sep 2026 Official leaderboard
22 GPT-5.4 Pro 1,777 Mar 2026 Official leaderboard
23 Claude Opus 4.8 1,768 May 2026 Official leaderboard
24 GLM-5.3-Flash 1,763 Aug 2026 Official leaderboard
25 Gemini 3.7 Flash 1,748 Aug 2026 Official leaderboard
26 GLM-5.3 1,740 Aug 2026 Official leaderboard
27 Muse Spark 1.3 1,730 Sep 2026 Official leaderboard
28 GPT 6 Luna Pro 1,728 Official leaderboard
29 Kimi K3 1,697 Jul 2026 Official leaderboard
30 GPT 5.6 Luna Pro 1,686 Official leaderboard
31 Gemini 3.1 Pro 1,661 Feb 2026 Official leaderboard
32 GPT-5.4 1,651 Mar 2026 Official leaderboard
33 Gemini 3.5 Flash 1,636 May 2026 Official leaderboard
34 Claude Sonnet 5 1,610 Jun 2026 Official leaderboard
35 Gemini 3.6 Flash 1,603 Jul 2026 Official leaderboard
36 Claude Opus 4.7 1,593 Apr 2026 Official leaderboard
37 Claude Haiku 5.5 1,565 Oct 2026 Official leaderboard
38 GPT-5.3 Codex 1,564 Feb 2026 Official leaderboard
39 Muse Spark 1.2 1,552 Aug 2026 Official leaderboard
40 GLM-5.2 1,547 Jun 2026 Official leaderboard
41 GPT-5.2 Pro 1,539 Dec 2025 Official leaderboard
42 Grok 4.5 1,538 Jul 2026 Official leaderboard
43 GLM-5.1 1,501 Apr 2026 Official leaderboard
44 DeepSeek V4 Flash 0731 1,482 Jul 2026 Official leaderboard
45 Claude Opus 4.6 1,472 Feb 2026 Official leaderboard
46 GPT-5.2 1,456 Dec 2025 Official leaderboard
47 DeepSeek-V4-Pro 1,442 Apr 2026 Official leaderboard
48 Kimi K2.6 1,425 Apr 2026 Official leaderboard
49 GPT-5.4 Mini 1,387 Mar 2026 Official leaderboard
50 Gemini 3 Pro 1,375 Nov 2025 Official leaderboard
51 Claude Sonnet 4.6 1,345 Feb 2026 Official leaderboard
52 Gemini 3.5 Flash-Lite 1,339 Jul 2026 Official leaderboard
53 Claude Opus 4.5 1,308 Nov 2025 Official leaderboard
54 Grok 4.20 1,280 Feb 2026 Official leaderboard
55 Qwen3.5 397B-A17B 1,275 Feb 2026 Official leaderboard
56 GPT-5.4 Nano 1,248 Mar 2026 Official leaderboard
57 Gemini 3 Flash 1,244 Dec 2025 Official leaderboard
58 Qwen 3.8 Max 1,218 Aug 2026 Official leaderboard
59 GLM-5 1,214 Feb 2026 Official leaderboard
60 Gemma 4 31B IT 1,213 Apr 2026 Official leaderboard
61 MiniMax-M2.7 1,194 Mar 2026 Official leaderboard
62 Kimi K2.5 1,187 Jan 2026 Official leaderboard
63 Grok 4.3 1,147 Apr 2026 Official leaderboard
64 Claude Sonnet 4.5 1,140 Sep 2025 Official leaderboard
65 Gemini 2.5 Pro Jun 2025 1,128 Jun 2025 Official leaderboard
66 GLM-4.7 1,100 Dec 2025 Official leaderboard
67 GPT-5 mini 1,076 Aug 2025 Official leaderboard
68 MiniMax-M2.5 1,068 Feb 2026 Official leaderboard
69 GPT-5.2 Codex 1,052 Dec 2025 Official leaderboard
70 DeepSeek-V3.2 1,037 Dec 2025 Official leaderboard
71 GPT-4.1 1,033 Apr 2025 Official leaderboard
72 Qwen3-Max 990 Sep 2025 Official leaderboard
73 Kimi K2 Sep 2025 963 Sep 2025 Official leaderboard
74 Grok 4.1 Fast 953 Nov 2025 Official leaderboard
75 Gemini 3.1 Flash-Lite 944 Mar 2026 Official leaderboard
76 GPT 4.5 (web harness) 917 Official leaderboard
77 gpt-oss-120b 777 Aug 2025 Official leaderboard
78 GPT-4o Mar 2025 744 Mar 2025 Official leaderboard
79 Llama 4 Maverick 628 Apr 2025 Official leaderboard
80 GPT-5 nano 563 Aug 2025 Official leaderboard

The best score, 2,206, costs $2.31 a task

Runs on the cost frontier score higher than any run that's cheaper. Gemini 4 Argon comes within 34 of the best for $0.61 a task, 4 times cheaper. The costs are the leaderboard's own estimates, so compare them only within this benchmark.

DeepSeek Z.ai OpenAI Google DeepMind Anthropic Moonshot AI Cost frontier One model at each effort level
0 500 1,000 1,500 2,000 2,500 $0.01 $0.10 $1.00 $10 $100 Cost per task, log scale GPT-6 Astra Pro Gemini 4 Carbon Gemini 4 Hydrogen DeepSeek V4.1 Flash Claude Haiku 5.5 0 500 1,000 1,500 2,000 2,500 $0.01 $0.10 $1.00 $10 $100 Cost per task, log scale GPT-6 Astra Pro Gemini 4 Hydrogen

Source: Official MineBench leaderboard

ModelEffortHarnessScoreCost per taskOutput tokens
GPT-6 Astra Pro 2,206 $2.31
Gemini 4 Carbon 2,195 $0.78
Gemini 4 Argon 2,172 $0.61
Claude Opus 5.5 2,112 $7.43
Claude Fable 5.2 2,102 $2.54
Gemini 4 Hydrogen 1,993 $0.46
Claude Opus 5 1,988 $6.00
Claude Sonnet 5.5 1,977 $0.67
GPT-5.6 Sol Pro 1,973 $47
Kimi K3.1 Pro 1,970 $0.42
Claude Fable 5.1 1,909 $9.84
GPT-5.5 Pro 1,903 $15
Claude Fable 5 1,851 $3.66
Muse Wonder 1,850 $0.69
Grok 4.7 1,829 $2.83
GPT-5.5 1,828 $1.33
DeepSeek V4.1 Flash 1,804 $0.07
Grok 4.6 1,793 $0.75
Gemini 3.8 Flash 1,789 $0.08
GPT-5.4 Pro 1,777 $29
Claude Opus 4.8 1,768 $2.77
GLM-5.3-Flash 1,763 $0.05
Gemini 3.7 Flash 1,748 $0.10
GLM-5.3 1,740 $0.43
Muse Spark 1.3 1,730 $0.44
Kimi K3 1,697 $0.85
Gemini 3.1 Pro 1,661
GPT-5.4 1,651
Gemini 3.5 Flash 1,636 $0.86
Claude Sonnet 5 1,610 $6.28
Gemini 3.6 Flash 1,603 $0.21
Claude Opus 4.7 1,593
Claude Haiku 5.5 1,565 $0.04
GPT-5.3 Codex 1,564
Muse Spark 1.2 1,552 $0.17
GLM-5.2 1,547 $0.30
GPT-5.2 Pro 1,539
Grok 4.5 1,538 $0.11
GLM-5.1 1,501 $0.28
DeepSeek V4 Flash 0731 1,482 $0.02
Claude Opus 4.6 1,472
GPT-5.2 1,456
DeepSeek-V4-Pro 1,442 $0.26
Kimi K2.6 1,425 $0.16
GPT-5.4 Mini 1,387
Gemini 3 Pro 1,375
Claude Sonnet 4.6 1,345
Gemini 3.5 Flash-Lite 1,339 $0.03
Claude Opus 4.5 1,308
Grok 4.20 1,280 $1.23
Qwen3.5 397B-A17B 1,275
GPT-5.4 Nano 1,248
Gemini 3 Flash 1,244
Qwen 3.8 Max 1,218 $0.77
GLM-5 1,214
Gemma 4 31B IT 1,213
MiniMax-M2.7 1,194
Kimi K2.5 1,187
Grok 4.3 1,147 $0.06
Claude Sonnet 4.5 1,140
Gemini 2.5 Pro Jun 2025 1,128
GLM-4.7 1,100
GPT-5 mini 1,076
MiniMax-M2.5 1,068
GPT-5.2 Codex 1,052
DeepSeek-V3.2 1,037
GPT-4.1 1,033
Qwen3-Max 990
Kimi K2 Sep 2025 963
Grok 4.1 Fast 953
Gemini 3.1 Flash-Lite 944
gpt-oss-120b 777
GPT-4o Mar 2025 744
Llama 4 Maverick 628
GPT-5 nano 563

8 models have held number one on the official leaderboard since its release in January 2026

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

OpenAI Google DeepMind Anthropic
Own leaderboard 8 leaders
GPT-6 Astra Pro 2,206

Source: Official MineBench 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
500 1,000 1,500 2,000 2,500 2025 2026 2027 500 1,000 1,500 2,000 2,500 2025 2026 2027

Source: Official MineBench 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 Vision

All 281 in Vision
  • MMMU In use 11.5K college-level multimodal questions across 30 subjects. Multiple-choice and open accuracy. GPT-5.1 85.4%
  • MMMU-Pro In use Harder MMMU: college-level multimodal questions with more options and vision-only inputs to remove text shortcuts. GPT-5.6 Sol 84.6%
  • CharXiv In use Chart understanding on real arXiv figures. Reasoning questions require synthesizing information from complex charts. Claude Fable 5 93.5%
  • MATH-Vision (MathVision) In use 3,040 competition math problems with visual context. Answer accuracy. Qwen 3.8 Max 97.7%
  • MMStar In use Models answer 1,500 human-curated image questions that cannot be solved without the image, across 6 core capabilities. Scored by accuracy. Qwen 3.8 Max 85.9%
  • ChartQA In use Models answer human-written and generated questions about chart images that need visual and arithmetic reasoning. Scored by relaxed accuracy with a 5 percent numeric tolerance. Command A Vision 90.9%