Benchmark Almanac

Excel Modeling Benchmark

Claude Fable 5.1 76.7% leads since September 2026
Not in model cards

Agents build investment banking and private equity financial models in Excel from a written brief and source files. Rubrics give partial credit for tabs, structure, formulas and formatting.

103 expert questions split into 1 public task, 51 licensable validation and 51 private test tasks. Each runs in Template and Scratch modes. The v1.0 page is archived.

Made by
Vals AI
Released
July 2026
Kind
Finance
Level
Expert
Scored
Graded by a model
Answers
Agent task
Tasks
Kept private
Strongest model tested
GPT-6 Astra #3 on the capability index
Last model added
October 2026
Scores refreshed
November 12th, 2026
76
Models scored on its own leaderboard

Claude Fable 5.1 leads with 76.7%

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.

76 models. Scroll sideways for the rest.

  1. 76.7 Claude Fable 5.1
  2. 75.9 Claude Opus 5.5
  3. 75.7 Claude Sonnet 5.5
  4. 75.2 Gemini 4 Argon
  5. 73.7 Claude Fable 5
  6. 73.6 Claude Opus 5
  7. 72.3 GPT-5.6 Sol
  8. 72.2 Gemini 3.8 Flash
  9. 72.0 Qwen3.8 Max (0902)
  10. 71.9 Muse Wonder Full
  11. 71.7 GPT-6 Astra
  12. 71.5 GPT-6 Sol
  13. 71.3 Gemini 3.7 Flash
  14. 70.8 GPT-6.1 Sol
  15. 69.4 Claude Opus 4.8
  16. 68.5 GPT-6 Luna
  17. 67.4 Muse Spark 1.3
  18. 67.1 GPT-5.6 Luna
  19. 67.1 Muse Wonder
  20. 67.0 Grok 4.7
  21. 66.7 Kimi K3
  22. 66.3 Claude Sonnet 5
  23. 66.2 GPT-5.6 Terra
  24. 65.5 MiMo-V2.6 Flash
  25. 65.4 Gemini 3.6 Flash
  26. 64.5 GPT-5.5
  27. 63.7 Claude Opus 4.7
  28. 63.5 Gemini 3.5 Flash
  29. 63.4 Ember-1
  30. 62.9 MiMo-V2.6 Pro
  31. 62.7 Grok 4.6
  32. 61.8 DeepSeek V4.1 Flash 1017
  33. 61.5 GLM-5.2
  34. 60.1 Claude Sonnet 4.6
  35. 60.1 Qwen 3.8 Max
  36. 59.7 Qwen 3.8 27B
  37. 59.0 Claude Haiku 5.5
  38. 58.1 Tencent Hy4 preview
  39. 57.9 Kimi K2.6
  40. 57.2 DeepSeek V4.1 Flash
  41. 57.1 Step 5 Preview
  42. 57.0 Qwen3.7-Max
  43. 57.0 DeepSeek V4 Flash 0731
  44. 57.0 Muse Spark 1.2
  45. 56.4 Muse Spark 1.1
  46. 56.3 GLM-5.3
  47. 56.2 Mistral Large 4 Preview
  48. 55.9 GLM-5.3-Flash
  49. 55.2 MiMo-V2.5-Pro
  50. 55.1 MiMo-V2.5
  51. 53.7 Muse Spark Open
  52. 52.9 Grok 4.5
  53. 52.8 DeepSeek V4 Pro 0813
  54. 52.6 Gemini 3.1 Pro
  55. 51.6 DeepSeek-V4-Pro
  56. 49.3 Qwen3.7-Plus
  57. 47.8 MiniMax-M3
  58. 45.4 GPT-5.4 Mini
  59. 44.8 GPT-5.4 Nano
  60. 43.2 Gemini 3.5 Flash-Lite
  61. 38.4 Inkling
  62. 36.8 Ling-3.0-flash-Fin
  63. 32.9 Qwen 3.6 Plus
  64. 31.8 Inkling-Small
  65. 31.7 Nemotron 3 Ultra
  66. 28.5 Kimi K2.5
  67. 28.4 MiniMax-M2.7
  68. 26.0 Gemini 3 Flash
  69. 25.0 Ling-3.0-flash
  70. 22.7 Claude Haiku 4.5
  71. 18.4 Grok 4.3
  72. 17.0 Nemotron 3.5 Lightning 30B-A3B
  73. 11.7 Grok 4.20
  74. 11.2 Mistral Medium 3.5
  75. 8.79 Mercury 2.5
  76. 8.63 Gemini 3.1 Flash-Lite
  1. 1 Claude Fable 5.1 76.7%
  2. 2 Claude Opus 5.5 75.9%
  3. 3 Claude Sonnet 5.5 75.7%
  4. 4 Gemini 4 Argon 75.2%
  5. 5 Claude Fable 5 73.7%
  6. 6 Claude Opus 5 73.6%
  7. 7 GPT-5.6 Sol 72.3%
  8. 8 Gemini 3.8 Flash 72.2%
  9. 9 Qwen3.8 Max (0902) 72%
  10. 10 Muse Wonder Full 71.9%
  11. 11 GPT-6 Astra 71.7%
  12. 12 GPT-6 Sol 71.5%
  13. 13 Gemini 3.7 Flash 71.3%
  14. 14 GPT-6.1 Sol 70.8%
  15. 15 Claude Opus 4.8 69.4%
  16. 16 GPT-6 Luna 68.5%
  17. 17 Muse Spark 1.3 max 67.4%
  18. 18 GPT-5.6 Luna 67.1%
  19. 19 Muse Wonder 67.1%
  20. 20 Grok 4.7 67%
  21. 21 Kimi K3 66.7%
  22. 22 Claude Sonnet 5 66.3%
  23. 23 GPT-5.6 Terra 66.2%
  24. 24 MiMo-V2.6 Flash 65.5%
  25. 25 Gemini 3.6 Flash 65.4%
  26. 26 GPT-5.5 64.5%
  27. 27 Claude Opus 4.7 63.7%
  28. 28 Gemini 3.5 Flash 63.5%
  29. 29 Ember-1 63.4%
  30. 30 MiMo-V2.6 Pro 62.9%
  31. 31 Grok 4.6 62.7%
  32. 32 DeepSeek V4.1 Flash 1017 61.8%
  33. 33 GLM-5.2 61.5%
  34. 34 Claude Sonnet 4.6 60.1%
  35. 35 Qwen 3.8 Max 60.1%
  36. 36 Qwen 3.8 27B 59.7%
  37. 37 Claude Haiku 5.5 59%
  38. 38 Tencent Hy4 preview 58.1%
  39. 39 Kimi K2.6 57.9%
  40. 40 DeepSeek V4.1 Flash 57.2%
  41. 41 Step 5 Preview 57.1%
  42. 42 Qwen3.7-Max 57%
  43. 43 DeepSeek V4 Flash 0731 57%
  44. 44 Muse Spark 1.2 57%
  45. 45 Muse Spark 1.1 56.4%
  46. 46 GLM-5.3 56.3%
  47. 47 Mistral Large 4 Preview 56.2%
  48. 48 GLM-5.3-Flash 55.9%
  49. 49 MiMo-V2.5-Pro 55.2%
  50. 50 MiMo-V2.5 55.1%
  51. 51 Muse Spark Open 53.7%
  52. 52 Grok 4.5 52.9%
  53. 53 DeepSeek V4 Pro 0813 52.8%
  54. 54 Gemini 3.1 Pro 52.6%
  55. 55 DeepSeek-V4-Pro 51.6%
  56. 56 Qwen3.7-Plus 49.3%
  57. 57 MiniMax-M3 47.8%
  58. 58 GPT-5.4 Mini 45.4%
  59. 59 GPT-5.4 Nano 44.8%
  60. 60 Gemini 3.5 Flash-Lite 43.2%
  61. 61 Inkling 38.4%
  62. 62 Ling-3.0-flash-Fin 36.8%
  63. 63 Qwen 3.6 Plus 32.9%
  64. 64 Inkling-Small 31.8%
  65. 65 Nemotron 3 Ultra 31.7%
  66. 66 Kimi K2.5 thinking 28.5%
  67. 67 MiniMax-M2.7 28.4%
  68. 68 Gemini 3 Flash 26%
  69. 69 Ling-3.0-flash 25%
  70. 70 Claude Haiku 4.5 thinking 22.7%
  71. 71 Grok 4.3 18.4%
  72. 72 Nemotron 3.5 Lightning 30B-A3B 17%
  73. 73 Grok 4.20 reasoning 11.7%
  74. 74 Mistral Medium 3.5 11.2%
  75. 75 Mercury 2.5 8.79%
  76. 76 Gemini 3.1 Flash-Lite 8.63%

Source: Official Excel Modeling Benchmark leaderboard

#ModelScoreSettingDateSource
1 Claude Fable 5.1 76.7% Sep 2026 Official leaderboard
2 Claude Opus 5.5 75.9% Sep 2026 Official leaderboard
3 Claude Sonnet 5.5 75.7% Sep 2026 Official leaderboard
4 Gemini 4 Argon 75.2% Sep 2026 Official leaderboard
5 Claude Fable 5 73.7% Jun 2026 Official leaderboard
6 Claude Opus 5 73.6% Jul 2026 Official leaderboard
7 GPT-5.6 Sol 72.3% Jul 2026 Official leaderboard
8 Gemini 3.8 Flash 72.2% Sep 2026 Official leaderboard
9 Qwen3.8 Max (0902) 72% Sep 2026 Official leaderboard
10 Muse Wonder Full 71.9% Oct 2026 Official leaderboard
11 GPT-6 Astra 71.7% Sep 2026 Official leaderboard
12 GPT-6 Sol 71.5% Sep 2026 Official leaderboard
13 Gemini 3.7 Flash 71.3% Aug 2026 Official leaderboard
14 GPT-6.1 Sol 70.8% Sep 2026 Official leaderboard
15 Claude Opus 4.8 69.4% May 2026 Official leaderboard
16 GPT-6 Luna 68.5% Sep 2026 Official leaderboard
17 Muse Spark 1.3 67.4% max Sep 2026 Official leaderboard
18 GPT-5.6 Luna 67.1% Jul 2026 Official leaderboard
19 Muse Wonder 67.1% Oct 2026 Official leaderboard
20 Grok 4.7 67% Sep 2026 Official leaderboard
21 Kimi K3 66.7% Jul 2026 Official leaderboard
22 Claude Sonnet 5 66.3% Jun 2026 Official leaderboard
23 GPT-5.6 Terra 66.2% Jul 2026 Official leaderboard
24 MiMo-V2.6 Flash 65.5% Sep 2026 Official leaderboard
25 Gemini 3.6 Flash 65.4% Jul 2026 Official leaderboard
26 GPT-5.5 64.5% Apr 2026 Official leaderboard
27 Claude Opus 4.7 63.7% Apr 2026 Official leaderboard
28 Gemini 3.5 Flash 63.5% May 2026 Official leaderboard
29 Ember-1 63.4% Sep 2026 Official leaderboard
30 MiMo-V2.6 Pro 62.9% Sep 2026 Official leaderboard
31 Grok 4.6 62.7% Aug 2026 Official leaderboard
32 DeepSeek V4.1 Flash 1017 61.8% Oct 2026 Official leaderboard
33 GLM-5.2 61.5% Jun 2026 Official leaderboard
34 Claude Sonnet 4.6 60.1% Feb 2026 Official leaderboard
35 Qwen 3.8 Max 60.1% Aug 2026 Official leaderboard
36 Qwen 3.8 27B 59.7% Aug 2026 Official leaderboard
37 Claude Haiku 5.5 59% Oct 2026 Official leaderboard
38 Tencent Hy4 preview 58.1% Aug 2026 Official leaderboard
39 Kimi K2.6 57.9% Apr 2026 Official leaderboard
40 DeepSeek V4.1 Flash 57.2% Sep 2026 Official leaderboard
41 Step 5 Preview 57.1% Sep 2026 Official leaderboard
42 Qwen3.7-Max 57% May 2026 Official leaderboard
43 DeepSeek V4 Flash 0731 57% Jul 2026 Official leaderboard
44 Muse Spark 1.2 57% Aug 2026 Official leaderboard
45 Muse Spark 1.1 56.4% Jul 2026 Official leaderboard
46 GLM-5.3 56.3% Aug 2026 Official leaderboard
47 Mistral Large 4 Preview 56.2% Oct 2026 Official leaderboard
48 GLM-5.3-Flash 55.9% Aug 2026 Official leaderboard
49 MiMo-V2.5-Pro 55.2% Apr 2026 Official leaderboard
50 MiMo-V2.5 55.1% Apr 2026 Official leaderboard
51 Muse Spark Open 53.7% Oct 2026 Official leaderboard
52 Grok 4.5 52.9% Jul 2026 Official leaderboard
53 DeepSeek V4 Pro 0813 52.8% Aug 2026 Official leaderboard
54 Gemini 3.1 Pro 52.6% Feb 2026 Official leaderboard
55 DeepSeek-V4-Pro 51.6% Apr 2026 Official leaderboard
56 Qwen3.7-Plus 49.3% Jun 2026 Official leaderboard
57 MiniMax-M3 47.8% Jun 2026 Official leaderboard
58 GPT-5.4 Mini 45.4% Mar 2026 Official leaderboard
59 GPT-5.4 Nano 44.8% Mar 2026 Official leaderboard
60 Gemini 3.5 Flash-Lite 43.2% Jul 2026 Official leaderboard
61 Inkling 38.4% Jul 2026 Official leaderboard
62 Ling-3.0-flash-Fin 36.8% Sep 2026 Official leaderboard
63 Qwen 3.6 Plus 32.9% Mar 2026 Official leaderboard
64 Inkling-Small 31.8% Jul 2026 Official leaderboard
65 Nemotron 3 Ultra 31.7% Jun 2026 Official leaderboard
66 Kimi K2.5 28.5% thinking Jan 2026 Official leaderboard
67 MiniMax-M2.7 28.4% Mar 2026 Official leaderboard
68 Gemini 3 Flash 26% Dec 2025 Official leaderboard
69 Ling-3.0-flash 25% Aug 2026 Official leaderboard
70 Claude Haiku 4.5 22.7% thinking Oct 2025 Official leaderboard
71 Grok 4.3 18.4% Apr 2026 Official leaderboard
72 Nemotron 3.5 Lightning 30B-A3B 17% Aug 2026 Official leaderboard
73 Grok 4.20 11.7% reasoning Feb 2026 Official leaderboard
74 Mistral Medium 3.5 11.2% Apr 2026 Official leaderboard
75 Mercury 2.5 8.79% Sep 2026 Official leaderboard
76 Gemini 3.1 Flash-Lite 8.63% Mar 2026 Official leaderboard

The best score, 76.7%, costs $16 a task

Runs on the cost frontier score higher than any run that's cheaper. Qwen3.8 Max (0902) comes within 4.7 points of the best for $4.43 a task, 4 times cheaper. The costs are the leaderboard's own estimates, so compare them only within this benchmark.

Google DeepMind Xiaomi Corp OpenAI Anthropic Qwen Cost frontier One model at each effort level
0% 20% 40% 60% 80% 100% $0.01 $0.10 $1.00 $10 $100 Cost per task, log scale Claude Fable 5.1 GPT-5.6 Sol GPT-6 Sol GPT-6 Luna MiMo-V2.5 Gemini 3.1 Flash-Lite 0% 20% 40% 60% 80% 100% $0.01 $0.10 $1.00 $10 $100 Cost per task, log scale Claude Fable 5.1 GPT-6.1 Sol

Source: Official Excel Modeling Benchmark leaderboard

ModelEffortHarnessScoreCost per taskOutput tokens
Claude Fable 5.1 76.7% $16 122,541
Claude Opus 5.5 75.9% $9.33 232,398
Claude Sonnet 5.5 75.7% $6.53 287,535
Gemini 4 Argon 75.2% $12 139,780
Claude Fable 5 73.7% $12 117,187
Claude Opus 5 73.6% $8.02 120,821
GPT-5.6 Sol 72.3% $6.01 53,506
Gemini 3.8 Flash 72.2% $8.24 212,969
Qwen3.8 Max (0902) 72% $4.43 142,004
Muse Wonder Full 71.9% $4.58 81,668
GPT-6 Astra 71.7% $5.83 54,340
GPT-6 Sol 71.5% $1.28 51,388
Gemini 3.7 Flash 71.3% $5.69 182,416
GPT-6.1 Sol 70.8% $1.14 59,382
Claude Opus 4.8 69.4% $12 217,025
GPT-6 Luna 68.5% $0.14 107,914
Muse Spark 1.3 max 67.4% $2.55 130,806
GPT-5.6 Luna 67.1% $0.37 68,190
Muse Wonder 67.1% $5.60 79,695
Grok 4.7 67% $6.48 152,252
Kimi K3 66.7% $3.07 95,082
Claude Sonnet 5 66.3% $10 254,896
GPT-5.6 Terra 66.2% $2.19 68,868
MiMo-V2.6 Flash 65.5% $0.11 157,412
Gemini 3.6 Flash 65.4% $3.89 149,409
GPT-5.5 64.5% $3.27 50,794
Claude Opus 4.7 63.7% $13 180,961
Gemini 3.5 Flash 63.5% $5.17 166,517
Ember-1 63.4% $2.46 71,047
MiMo-V2.6 Pro 62.9% $0.33 172,106
Grok 4.6 62.7% $3.06 178,805
Muse Spark 1.3 62.7% $3.55 121,547
DeepSeek V4.1 Flash 1017 61.8% $0.36 163,318
GLM-5.2 61.5% $4.47 190,787
Claude Sonnet 4.6 60.1% $7.23 206,994
Qwen 3.8 Max 60.1% $2.66 118,149
Qwen 3.8 27B 59.7% $4.56 176,776
Claude Haiku 5.5 59% $1.15 210,084
Tencent Hy4 preview 58.1% $0.79 102,269
Kimi K2.6 57.9% $2.20 143,522
DeepSeek V4.1 Flash 57.2% $0.23 130,189
Step 5 Preview 57.1% $1.33 144,151
Qwen3.7-Max 57% $2.86 90,783
DeepSeek V4 Flash 0731 57% $0.38 148,388
Muse Spark 1.2 57% $2.31 105,305
Muse Spark 1.1 56.4% $1.11 88,917
GLM-5.3 56.3% $3.79 178,213
Mistral Large 4 Preview 56.2% $4.84 170,078
GLM-5.3-Flash 55.9% $0.71 286,761
MiMo-V2.5-Pro 55.2% $0.22 116,949
MiMo-V2.5 55.1% $0.08 122,702
Muse Spark Open 53.7% $5.35 83,293
Grok 4.5 52.9% $1.55 72,905
DeepSeek V4 Pro 0813 52.8% $1.08 149,550
Gemini 3.1 Pro 52.6% $3.77 53,359
DeepSeek-V4-Pro 51.6% $0.80 96,109
Qwen3.7-Plus 49.3% $0.60 107,428
MiniMax-M3 47.8% $2.09 134,398
GPT-5.4 Mini 45.4% $2.15 258,437
GPT-5.4 Nano 44.8% $0.57 160,916
Gemini 3.5 Flash-Lite 43.2% $1.23 105,076
Inkling 38.4% $1.67 90,997
Ling-3.0-flash-Fin 36.8% $0.18 162,482
Qwen 3.6 Plus 32.9% $2.87 83,889
Inkling-Small 31.8% $0.25 62,374
Nemotron 3 Ultra 31.7% 125,620
Kimi K2.5 thinking 28.5% $0.79 66,862
MiniMax-M2.7 28.4% $1.66 84,099
Gemini 3 Flash 26% $0.45 47,475
Ling-3.0-flash 25% $0.18 182,208
Claude Haiku 4.5 thinking 22.7% $1.03 80,653
Grok 4.3 18.4% $0.63 45,001
Nemotron 3.5 Lightning 30B-A3B 17% $0.18 295,087
Grok 4.20 reasoning 11.7% $0.47 18,230
Mistral Medium 3.5 11.2% $11 61,464
Mercury 2.5 8.79% $0.42 54,229
Gemini 3.1 Flash-Lite 8.63% $0.07 11,640

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.

Anthropic
Own leaderboard 2 leaders
Claude Fable 5.1 76.7%

Source: Official Excel Modeling Benchmark 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 Best so far
0% 20% 40% 60% 80% 100% 2026 Near the ceiling (85%) Claude Fable 5.1 Claude Opus 4.8 GPT-5.5 Claude Sonnet 4.6 Kimi K2.5 Claude Haiku 4.5 0% 20% 40% 60% 80% 100% 2026 Near the ceiling (85%) Claude Fable 5.1 Claude Opus 4.8

Source: Official Excel Modeling Benchmark 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.

Versions

We switch the top row to each new version when it comes out. Open it for one row per version. An old version stays in play while people still test it, so whoever first takes it past 85% gets credit for closing it out.

Anthropic
Excel Modeling Benchmark v1.1 4 leaders
Claude Fable 5.2 77.4%
Excel Modeling Benchmark 2 leaders
Claude Fable 5.1 76.7%
Excel Modeling Benchmark v1.1 2 leaders
Claude Fable 5.2 77.4%

Source: Official leaderboards

Replaced by Excel Modeling Benchmark v1.1

More in Law, finance and medicine

All 179 in Law, finance and medicine
  • MedXpertQA-MM In use Models answer expert-level medical exam questions that combine images with clinical records and test results. Scoring is accuracy. Qwen 3.8 Max 80.4%
  • HealthBench Professional In use 525 physician-authored tasks from real clinician chats (care consult, documentation, research) with physician-adjudicated rubrics. Claude Sonnet 5.5 69.2%
  • HealthBench In use 5,000 realistic health conversations graded against physician-written rubrics. A length-adjusted variant penalizes verbosity. GPT-5 67.2%
  • FinSearchComp In use Agents search the web to fetch live financial data, look up history, and run multi-step investigations, across 635 expert-written questions. Scored by answer accuracy. Ling-3.1-flash 81.7%
  • KBL In use Models answer Korean legal knowledge and reasoning tasks and Korean bar exam questions. It has 7 knowledge tasks, 4 reasoning tasks and 2,510 bar exam items. Solar Pro 4 77.5%
  • KorMedMCQA In use Models answer 7,469 multiple-choice questions from Korean licensing exams for doctors, nurses, pharmacists and dentists, held from 2012 to 2024. Solar Pro 4 93.2%