Benchmark Almanac

Excel Modeling Benchmark v1.1

Claude Fable 5.2 77.4% leads since November 2026
In use

Agents build the same 103 Excel financial modeling tasks as EMB v1.0, with the Scratch rubric judge changed to GPT-6 Luna.

Vals says v1.1 scores are not directly comparable with v1.0. Part of Vals Index v2.2 from October 19th, 2026. The board had 37 models on October 29th, 2026.

Made by
Vals AI
Released
October 2026
Kind
Finance
Level
Expert
Scored
Graded by a model
Answers
Agent task
Tasks
Kept private
Strongest model tested
Claude Fable 5.2 #1 on the capability index
Last in a launch post
November 2026
Last model added
November 2026
Scores refreshed
November 13th, 2026
42
Models scored on its own leaderboard
1
Launch post lists it, from 1 lab

Claude Fable 5.2 leads with 77.4%

Best score per model 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.

42 models. Scroll sideways for the rest.

  1. 77.4 Claude Fable 5.2
  2. 76.7 Claude Fable 5.1
  3. 76.5 Gemini 4 Carbon
  4. 75.9 Claude Sonnet 5.5
  5. 75.8 Claude Opus 5.5
  6. 75.8 Gemini 4 Hydrogen
  7. 75.0 Muse Wonder Full
  8. 74.8 Gemini 4 Argon
  9. 73.3 GPT-6.1 Terra
  10. 72.9 Gemini 3.8 Flash
  11. 72.7 GPT-6 Astra
  12. 72.5 Muse Wonder
  13. 72.2 Muse Ember
  14. 72.0 Qwen3.8 Max (0902)
  15. 71.5 Grok 4.7
  16. 71.3 Kimi K3
  17. 71.1 GPT-6.1 Sol
  18. 70.0 Kimi K3.1 Pro
  19. 70.0 GPT-6 Luna
  20. 69.5 Step 5 Preview
  21. 69.5 Inkling 2 Small
  22. 69.2 DeepSeek V4.1 Flash 1017
  23. 69.1 GLM-5.3
  24. 69.0 Mistral Large 4 Preview
  25. 69.0 Tencent Hy4 preview
  26. 68.9 MiMo-V2.6 Flash
  27. 68.7 Mistral Large 4
  28. 67.8 MiniMax-M3.1-Flash
  29. 67.8 DeepSeek V4.1 Flash
  30. 67.3 MiMo-V2.6 Pro
  31. 66.3 GLM-5.3-Flash
  32. 66.2 Tencent Hy4
  33. 66.0 Qwen 3.8 Max
  34. 65.5 Claude Haiku 5.5
  35. 65.1 DeepSeek V4 Pro 0813
  36. 64.7 Beam 1.0
  37. 62.0 Muse Spark Open
  38. 59.8 MiniMax-M3
  39. 57.9 Mercury 3
  40. 52.8 Nemotron 3.5 Super VL
  41. 41.0 Inkling
  42. 40.8 Inkling-Small
  1. 1 Claude Fable 5.2 77.4%
  2. 2 Claude Fable 5.1 76.7%
  3. 3 Gemini 4 Carbon 76.5%
  4. 4 Claude Sonnet 5.5 75.9%
  5. 5 Claude Opus 5.5 75.8%
  6. 6 Gemini 4 Hydrogen 75.8%
  7. 7 Muse Wonder Full 75%
  8. 8 Gemini 4 Argon 74.8%
  9. 9 GPT-6.1 Terra 73.3%
  10. 10 Gemini 3.8 Flash 72.9%
  11. 11 GPT-6 Astra 72.7%
  12. 12 Muse Wonder 72.5%
  13. 13 Muse Ember 72.2%
  14. 14 Qwen3.8 Max (0902) 72%
  15. 15 Grok 4.7 71.5%
  16. 16 Kimi K3 71.3%
  17. 17 GPT-6.1 Sol 71.1%
  18. 18 Kimi K3.1 Pro 70%
  19. 19 GPT-6 Luna 70%
  20. 20 Step 5 Preview 69.5%
  21. 21 Inkling 2 Small 69.5%
  22. 22 DeepSeek V4.1 Flash 1017 69.2%
  23. 23 GLM-5.3 69.1%
  24. 24 Mistral Large 4 Preview 69%
  25. 25 Tencent Hy4 preview 69%
  26. 26 MiMo-V2.6 Flash 68.9%
  27. 27 Mistral Large 4 68.7%
  28. 28 MiniMax-M3.1-Flash 67.8%
  29. 29 DeepSeek V4.1 Flash 67.8%
  30. 30 MiMo-V2.6 Pro 67.3%
  31. 31 GLM-5.3-Flash 66.3%
  32. 32 Tencent Hy4 66.2%
  33. 33 Qwen 3.8 Max 66%
  34. 34 Claude Haiku 5.5 65.5%
  35. 35 DeepSeek V4 Pro 0813 65.1%
  36. 36 Beam 1.0 64.7%
  37. 37 Muse Spark Open 62%
  38. 38 MiniMax-M3 59.8%
  39. 39 Mercury 3 57.9%
  40. 40 Nemotron 3.5 Super VL 52.8%
  41. 41 Inkling 41%
  42. 42 Inkling-Small 40.8%

Source: Official Excel Modeling Benchmark v1.1 leaderboard

#ModelScoreSettingDateSource
1 Claude Fable 5.2 77.4% Nov 2026 Official leaderboard
2 Claude Fable 5.1 76.7% Sep 2026 Official leaderboard
3 Gemini 4 Carbon 76.5% Oct 2026 Official leaderboard
4 Claude Sonnet 5.5 75.9% Sep 2026 Official leaderboard
5 Claude Opus 5.5 75.8% Sep 2026 Official leaderboard
6 Gemini 4 Hydrogen 75.8% Nov 2026 Official leaderboard
7 Muse Wonder Full 75% Oct 2026 Official leaderboard
8 Gemini 4 Argon 74.8% Sep 2026 Official leaderboard
9 GPT-6.1 Terra 73.3% Oct 2026 Official leaderboard
10 Gemini 3.8 Flash 72.9% Sep 2026 Official leaderboard
11 GPT-6 Astra 72.7% Sep 2026 Official leaderboard
12 Muse Wonder 72.5% Oct 2026 Official leaderboard
13 Muse Ember 72.2% Nov 2026 Official leaderboard
14 Qwen3.8 Max (0902) 72% Sep 2026 Official leaderboard
15 Grok 4.7 71.5% Sep 2026 Official leaderboard
16 Kimi K3 71.3% Jul 2026 Official leaderboard
17 GPT-6.1 Sol 71.1% Sep 2026 Official leaderboard
18 Kimi K3.1 Pro 70% Oct 2026 Official leaderboard
19 GPT-6 Luna 70% Sep 2026 Official leaderboard
20 Step 5 Preview 69.5% Sep 2026 Official leaderboard
21 Inkling 2 Small 69.5% Nov 2026 Official leaderboard
22 DeepSeek V4.1 Flash 1017 69.2% Oct 2026 Official leaderboard
23 GLM-5.3 69.1% Aug 2026 Official leaderboard
24 Mistral Large 4 Preview 69% Oct 2026 Official leaderboard
25 Tencent Hy4 preview 69% Aug 2026 Official leaderboard
26 MiMo-V2.6 Flash 68.9% Sep 2026 Official leaderboard
27 Mistral Large 4 68.7% Oct 2026 Official leaderboard
28 MiniMax-M3.1-Flash 67.8% Oct 2026 Official leaderboard
29 DeepSeek V4.1 Flash 67.8% Sep 2026 Official leaderboard
30 MiMo-V2.6 Pro 67.3% Sep 2026 Official leaderboard
31 GLM-5.3-Flash 66.3% Aug 2026 Official leaderboard
32 Tencent Hy4 66.2% Nov 2026 Official leaderboard
33 Qwen 3.8 Max 66% Aug 2026 Official leaderboard
34 Claude Haiku 5.5 65.5% Oct 2026 Official leaderboard
35 DeepSeek V4 Pro 0813 65.1% Aug 2026 Official leaderboard
36 Beam 1.0 64.7% Oct 2026 Official leaderboard
37 Muse Spark Open 62% Oct 2026 Official leaderboard
38 MiniMax-M3 59.8% Jun 2026 Official leaderboard
39 Mercury 3 57.9% Oct 2026 Official leaderboard
40 Nemotron 3.5 Super VL 52.8% Oct 2026 Official leaderboard
41 Inkling 41% Jul 2026 Official leaderboard
42 Inkling-Small 40.8% Jul 2026 Official leaderboard

Labs report scores for 1 model in their cards. We plot every source in the charts below.

Accuracy against latency

The Vals scatter chart with its x axis switched from Cost to Latency: mean time per test against accuracy, on a log scale. The line joins the runs that no other run beats on both measures. We redrew this chart from the numbers on the official leaderboard.

40% 60% 80% 100% 1001,00010K Latency (mean per test) (s), log scale, lower is better Accuracy MiniMax-M3: Accuracy 59.8%, latency (mean per test) 1,299 s Inkling: Accuracy 41%, latency (mean per test) 1,582 s Inkling-Small: Accuracy 40.8%, latency (mean per test) 1,174 s Kimi K3: Accuracy 71.3%, latency (mean per test) 741 s Qwen 3.8 Max: Accuracy 66%, latency (mean per test) 3,467 s DeepSeek V4 Pro 0813: Accuracy 65.1%, latency (mean per test) 2,247 s GLM-5.3: Accuracy 69.1%, latency (mean per test) 2,172 s GLM-5.3-Flash: Accuracy 66.3%, latency (mean per test) 4,495 s Tencent Hy4 preview: Accuracy 69%, latency (mean per test) 1,839 s Claude Fable 5.1: Accuracy 76.7%, latency (mean per test) 1,772 s Qwen3.8 Max (0902): Accuracy 72%, latency (mean per test) 3,193 s Gemini 3.8 Flash: Accuracy 72.9%, latency (mean per test) 769 s GPT-6 Astra: Accuracy 72.7%, latency (mean per test) 804 s DeepSeek V4.1 Flash: Accuracy 67.8%, latency (mean per test) 780 s Step 5 Preview: Accuracy 69.5%, latency (mean per test) 2,615 s Grok 4.7: Accuracy 71.5%, latency (mean per test) 2,097 s MiMo-V2.6 Flash: Accuracy 68.9%, latency (mean per test) 1,522 s MiMo-V2.6 Pro: Accuracy 67.3%, latency (mean per test) 3,314 s Claude Opus 5.5: Accuracy 75.8%, latency (mean per test) 2,106 s GPT-6 Luna: Accuracy 70%, latency (mean per test) 1,041 s Claude Sonnet 5.5: Accuracy 75.9%, latency (mean per test) 2,205 s GPT-6.1 Sol: Accuracy 71.1%, latency (mean per test) 1,108 s Gemini 4 Argon: Accuracy 74.8%, latency (mean per test) 2,306 s Beam 1.0: Accuracy 64.7%, latency (mean per test) 1,204 s Mistral Large 4: Accuracy 68.7%, latency (mean per test) 1,387 s Mistral Large 4 Preview: Accuracy 69%, latency (mean per test) 4,905 s Claude Haiku 5.5: Accuracy 65.5%, latency (mean per test) 1,196 s Muse Wonder: Accuracy 72.5%, latency (mean per test) 1,034 s Muse Wonder Full: Accuracy 75%, latency (mean per test) 1,685 s Muse Spark Open: Accuracy 62%, latency (mean per test) 5,079 s Nemotron 3.5 Super VL: Accuracy 52.8%, latency (mean per test) 817 s DeepSeek V4.1 Flash 1017: Accuracy 69.2%, latency (mean per test) 1,972 s Mercury 3: Accuracy 57.9%, latency (mean per test) 531 s MiniMax-M3.1-Flash: Accuracy 67.8%, latency (mean per test) 1,848 s Kimi K3.1 Pro: Accuracy 70%, latency (mean per test) 1,283 s GPT-6.1 Terra: Accuracy 73.3%, latency (mean per test) 757 s Gemini 4 Carbon: Accuracy 76.5%, latency (mean per test) 1,649 s Tencent Hy4: Accuracy 66.2%, latency (mean per test) 1,059 s Inkling 2 Small: Accuracy 69.5%, latency (mean per test) 834 s Claude Fable 5.2: Accuracy 77.4%, latency (mean per test) 979 s Muse Ember: Accuracy 72.2%, latency (mean per test) 290 s Gemini 4 Hydrogen: Accuracy 75.8%, latency (mean per test) 663 s Claude Fable 5.2Gemini 4 HydrogenMuse Ember 40% 60% 80% 100% 1001,00010K Latency (mean per test) (s), log scale, lower is better Accuracy MiniMax-M3: Accuracy 59.8%, latency (mean per test) 1,299 s Inkling: Accuracy 41%, latency (mean per test) 1,582 s Inkling-Small: Accuracy 40.8%, latency (mean per test) 1,174 s Kimi K3: Accuracy 71.3%, latency (mean per test) 741 s Qwen 3.8 Max: Accuracy 66%, latency (mean per test) 3,467 s DeepSeek V4 Pro 0813: Accuracy 65.1%, latency (mean per test) 2,247 s GLM-5.3: Accuracy 69.1%, latency (mean per test) 2,172 s GLM-5.3-Flash: Accuracy 66.3%, latency (mean per test) 4,495 s Tencent Hy4 preview: Accuracy 69%, latency (mean per test) 1,839 s Claude Fable 5.1: Accuracy 76.7%, latency (mean per test) 1,772 s Qwen3.8 Max (0902): Accuracy 72%, latency (mean per test) 3,193 s Gemini 3.8 Flash: Accuracy 72.9%, latency (mean per test) 769 s GPT-6 Astra: Accuracy 72.7%, latency (mean per test) 804 s DeepSeek V4.1 Flash: Accuracy 67.8%, latency (mean per test) 780 s Step 5 Preview: Accuracy 69.5%, latency (mean per test) 2,615 s Grok 4.7: Accuracy 71.5%, latency (mean per test) 2,097 s MiMo-V2.6 Flash: Accuracy 68.9%, latency (mean per test) 1,522 s MiMo-V2.6 Pro: Accuracy 67.3%, latency (mean per test) 3,314 s Claude Opus 5.5: Accuracy 75.8%, latency (mean per test) 2,106 s GPT-6 Luna: Accuracy 70%, latency (mean per test) 1,041 s Claude Sonnet 5.5: Accuracy 75.9%, latency (mean per test) 2,205 s GPT-6.1 Sol: Accuracy 71.1%, latency (mean per test) 1,108 s Gemini 4 Argon: Accuracy 74.8%, latency (mean per test) 2,306 s Beam 1.0: Accuracy 64.7%, latency (mean per test) 1,204 s Mistral Large 4: Accuracy 68.7%, latency (mean per test) 1,387 s Mistral Large 4 Preview: Accuracy 69%, latency (mean per test) 4,905 s Claude Haiku 5.5: Accuracy 65.5%, latency (mean per test) 1,196 s Muse Wonder: Accuracy 72.5%, latency (mean per test) 1,034 s Muse Wonder Full: Accuracy 75%, latency (mean per test) 1,685 s Muse Spark Open: Accuracy 62%, latency (mean per test) 5,079 s Nemotron 3.5 Super VL: Accuracy 52.8%, latency (mean per test) 817 s DeepSeek V4.1 Flash 1017: Accuracy 69.2%, latency (mean per test) 1,972 s Mercury 3: Accuracy 57.9%, latency (mean per test) 531 s MiniMax-M3.1-Flash: Accuracy 67.8%, latency (mean per test) 1,848 s Kimi K3.1 Pro: Accuracy 70%, latency (mean per test) 1,283 s GPT-6.1 Terra: Accuracy 73.3%, latency (mean per test) 757 s Gemini 4 Carbon: Accuracy 76.5%, latency (mean per test) 1,649 s Tencent Hy4: Accuracy 66.2%, latency (mean per test) 1,059 s Inkling 2 Small: Accuracy 69.5%, latency (mean per test) 834 s Claude Fable 5.2: Accuracy 77.4%, latency (mean per test) 979 s Muse Ember: Accuracy 72.2%, latency (mean per test) 290 s Gemini 4 Hydrogen: Accuracy 75.8%, latency (mean per test) 663 s Claude Fable 5.2Gemini 4 HydrogenMuse Ember
ModelSettingAccuracyLatency (mean per test)
Claude Fable 5.2 77.4% 979.09 s
Claude Fable 5.1 76.7% 1,772 s
Gemini 4 Carbon 76.5% 1,649 s
Claude Sonnet 5.5 75.9% 2,205 s
Claude Opus 5.5 75.8% 2,106 s
Gemini 4 Hydrogen 75.8% 662.89 s
Muse Wonder Full 75% 1,685 s
Gemini 4 Argon 74.8% 2,306 s
GPT-6.1 Terra 73.3% 756.89 s
Gemini 3.8 Flash 72.9% 769.2 s
GPT-6 Astra 72.7% 804.28 s
Muse Wonder 72.5% 1,034 s
Muse Ember 72.2% 289.84 s
Qwen3.8 Max (0902) 72% 3,193 s
Grok 4.7 71.5% 2,097 s
Kimi K3 71.3% 741.42 s
GPT-6.1 Sol 71.1% 1,108 s
Kimi K3.1 Pro 70% 1,283 s
GPT-6 Luna 70% 1,041 s
Step 5 Preview 69.6% 2,615 s
Inkling 2 Small 69.5% 833.92 s
DeepSeek V4.1 Flash 1017 69.2% 1,972 s
GLM-5.3 69.1% 2,172 s
Mistral Large 4 Preview 69% 4,905 s
Tencent Hy4 preview 69% 1,839 s
MiMo-V2.6 Flash 68.9% 1,522 s
Mistral Large 4 68.7% 1,387 s
MiniMax-M3.1-Flash 67.8% 1,848 s
DeepSeek V4.1 Flash 67.8% 780.46 s
MiMo-V2.6 Pro 67.4% 3,314 s
GLM-5.3-Flash 66.3% 4,495 s
Tencent Hy4 66.2% 1,059 s
Qwen 3.8 Max 66% 3,467 s
Claude Haiku 5.5 65.5% 1,196 s
DeepSeek V4 Pro 0813 65.1% 2,247 s
Beam 1.0 64.7% 1,204 s
Muse Spark Open 62% 5,079 s
MiniMax-M3 59.8% 1,299 s
Mercury 3 57.9% 531.28 s
Nemotron 3.5 Super VL 52.8% 816.73 s
Inkling 41% 1,582 s
Inkling-Small 40.8% 1,174 s

The best score, 77.4%, costs $6.67 a task

Runs on the cost frontier score higher than any run that's cheaper. GPT-6.1 Terra comes within 4 points of the best for $0.56 a task, 12 times cheaper. The costs are the leaderboard's own estimates, so compare them only within this benchmark.

OpenAI Google DeepMind Anthropic Meta 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 5.2 GPT-6.1 Terra Muse Ember 0% 20% 40% 60% 80% 100% $0.01 $0.10 $1.00 $10 $100 Cost per task, log scale Claude Fable 5.2 Muse Ember

Source: Official Excel Modeling Benchmark v1.1 leaderboard

ModelEffortHarnessScoreCost per taskOutput tokens
Claude Fable 5.2 77.4% $6.67
Claude Fable 5.1 76.7% $16
Gemini 4 Carbon 76.5% $5.76
Claude Sonnet 5.5 75.9% $6.53
Claude Opus 5.5 75.8% $9.33
Gemini 4 Hydrogen 75.8% $3.17
Muse Wonder Full 75% $4.58
Gemini 4 Argon 74.8% $12
GPT-6.1 Terra 73.3% $0.56
Gemini 3.8 Flash 72.9% $8.24
GPT-6 Astra 72.7% $5.83
Muse Wonder 72.5% $5.60
Muse Ember 72.2% $0.09
Qwen3.8 Max (0902) 72% $4.43
Grok 4.7 71.5% $6.48
Kimi K3 71.3% $3.07
GPT-6.1 Sol 71.1% $1.14
Kimi K3.1 Pro 70% $1.59
GPT-6 Luna 70% $0.14
Step 5 Preview 69.5% $1.33
Inkling 2 Small 69.5% $0.53
DeepSeek V4.1 Flash 1017 69.2% $0.33
GLM-5.3 69.1% $3.79
Mistral Large 4 Preview 69% $2.47
Tencent Hy4 preview 69% $0.79
MiMo-V2.6 Flash 68.9% $0.11
Mistral Large 4 68.7% $0.83
MiniMax-M3.1-Flash 67.8% $0.27
DeepSeek V4.1 Flash 67.8% $0.23
MiMo-V2.6 Pro 67.3% $0.33
GLM-5.3-Flash 66.3% $1.43
Tencent Hy4 66.2% $0.62
Qwen 3.8 Max 66% $2.66
Claude Haiku 5.5 65.5% $1.36
DeepSeek V4 Pro 0813 65.1% $1.08
Beam 1.0 64.7%
Muse Spark Open 62% $5.35
MiniMax-M3 59.8% $3.18
Mercury 3 57.9% $0.19
Nemotron 3.5 Super VL 52.8% $1.29
Inkling 41% $1.67
Inkling-Small 40.8% $0.25

2 models have held number one on the official leaderboard since its release in October 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 5.2 77.4%

Source: Official Excel Modeling Benchmark v1.1 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 Reported by the lab in its model card Best so far
0% 20% 40% 60% 80% 100% 2026 2027 Near the ceiling (85%) Claude Fable 5.2 Claude Fable 5.1 Kimi K3 MiniMax-M3 0% 20% 40% 60% 80% 100% 2026 2027 Near the ceiling (85%) Claude Fable 5.2 MiniMax-M3

Sources: Official Excel Modeling Benchmark v1.1 leaderboard and labs’ model cards

One lab put it in a launch post in the last six months

Launch posts from every lab that has listed it, one tick per post. Tall colored ticks include it, and short gray ones leave it out.

Thinking Machines
DateLaunch postScoreAs the post writes it
Nov 5, 2026 Inkling 2 Small 69.5 Excel Modeling Benchmark v1.1, 69.5%

One lab claim next to the leaderboard

Each score a lab reported for its own model, beside the same model's best score on the leaderboard. Gaps mostly come from different harnesses, effort levels and task subsets. 1 of them match, which usually means one side copied the other. Every benchmark.

Model Lab says Board says Gap
Inkling 2 Small 69.5% Excel Modeling Benchmark v1.1, 69.5% 69.5% Leaderboard Same

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

Replaces Excel Modeling Benchmark

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%