Vals AI

Excel Modeling Benchmark (EMB)

Agents build full Excel financial models of the kind used in investment banking and private equity, from a scenario plus source files. Categories: LBO, DCF, M&A, operating models, comps, dataroom summaries and misc.

Best reported

76.7

Claude Fable 5.1Anthropic

accuracy %

Official board, 29 Oct 2026

About this benchmark

Size
103 tasks (1 public, 51 private validation, 51 test)
Public split
1 task
Grading
Template mode checks exact cell matches against a gold model. Scratch mode uses a partial-credit rubric scored by an LLM judge (GPT-6 Luna). Workbooks are recalculated in Excel. Headline is the mean over 7 equally weighted categories.
Licence
proprietary
Submission
Not stated
Products allowed
Not stated
Latest version
v1.1
What it does not cover
No research or retrieval from live filings, and no judgement on investment decisions.

Results

10 models
Leaderboard

Each score shows who reported it: the benchmark's authors (quoted from their paper), the benchmark's official board, or Moneda (run with the benchmark's own grader and rules). We do not copy other aggregators.

Best value: GPT-6.1 Terra, 73.3 at $0.560 per question (cheapest within 5% of the top score)

Excel Modeling Benchmark (EMB): score vs cost

Score against cost per test. Up and to the left is better.

Updated 29 Oct 2026

Y: accuracy %X: Cost per test (USD)Showing 10 models
Excel Modeling Benchmark (EMB): score vs cost10 models. Frontier: GPT-6.1 Terra, Wonder Full, Gemini 4 Carbon, Claude Fable 5.1.7374757677$0$5$10$15EfficientInefficient↑ Best performance← Lowest costCost per test (USD)Claude Fable 5.1: 76.7 accuracy %, $15.93 per test (Vals.ai EMB v1.1 leaderboard, Cost / Test, 29 Oct 2026)Claude Fable 5.1Gemini 4 Carbon: 76.5 accuracy %, $5.76 per test (Vals.ai EMB v1.1 leaderboard, Cost / Test, 29 Oct 2026)Gemini 4 CarbonClaude Sonnet 5.5: 75.9 accuracy %, $8.79 per test (Vals.ai EMB v1.1 leaderboard, Cost / Test, 29 Oct 2026)Claude Sonnet 5.5Claude Opus 5.5: 75.8 accuracy %, $11.02 per test (Vals.ai EMB v1.1 leaderboard, Cost / Test, 29 Oct 2026)Claude Opus 5.5Wonder Full: 75 accuracy %, $4.58 per test (Vals.ai EMB v1.1 leaderboard, Cost / Test, 29 Oct 2026)Wonder FullGemini 4 Argon: 74.8 accuracy %, $11.8 per test (Vals.ai EMB v1.1 leaderboard, Cost / Test, 29 Oct 2026)Gemini 4 ArgonGPT-6.1 Terra: 73.3 accuracy %, $0.56 per test (Vals.ai EMB v1.1 leaderboard, Cost / Test, 29 Oct 2026) GPT-6.1 TerraGemini 3.8 Flash: 72.9 accuracy %, $8.24 per test (Vals.ai EMB v1.1 leaderboard, Cost / Test, 29 Oct 2026)Gemini 3.8 FlashGPT-6 Astra: 72.7 accuracy %, $5.83 per test (Vals.ai EMB v1.1 leaderboard, Cost / Test, 29 Oct 2026) GPT-6 AstraWonder: 72.5 accuracy %, $5.6 per test (Vals.ai EMB v1.1 leaderboard, Cost / Test, 29 Oct 2026)WonderExcel Modeling Benchmark (EMB): score vs cost10 models. Frontier: GPT-6.1 Terra, Wonder Full, Gemini 4 Carbon, Claude Fable 5.1.7374757677$0.5$1$2$5$10$20EfficientInefficient↑ Best performance← Lowest costCost per test (USD)Claude Fable 5.1: 76.7 accuracy %, $15.93 per test (Vals.ai EMB v1.1 leaderboard, Cost / Test, 29 Oct 2026)Claude Fable 5.1Gemini 4 Carbon: 76.5 accuracy %, $5.76 per test (Vals.ai EMB v1.1 leaderboard, Cost / Test, 29 Oct 2026)Gemini 4 CarbonClaude Sonnet 5.5: 75.9 accuracy %, $8.79 per test (Vals.ai EMB v1.1 leaderboard, Cost / Test, 29 Oct 2026)Claude Sonnet 5.5Claude Opus 5.5: 75.8 accuracy %, $11.02 per test (Vals.ai EMB v1.1 leaderboard, Cost / Test, 29 Oct 2026)Claude Opus 5.5Wonder Full: 75 accuracy %, $4.58 per test (Vals.ai EMB v1.1 leaderboard, Cost / Test, 29 Oct 2026)Wonder FullGemini 4 Argon: 74.8 accuracy %, $11.8 per test (Vals.ai EMB v1.1 leaderboard, Cost / Test, 29 Oct 2026)Gemini 4 ArgonGPT-6.1 Terra: 73.3 accuracy %, $0.56 per test (Vals.ai EMB v1.1 leaderboard, Cost / Test, 29 Oct 2026) GPT-6.1 TerraGemini 3.8 Flash: 72.9 accuracy %, $8.24 per test (Vals.ai EMB v1.1 leaderboard, Cost / Test, 29 Oct 2026)Gemini 3.8 FlashGPT-6 Astra: 72.7 accuracy %, $5.83 per test (Vals.ai EMB v1.1 leaderboard, Cost / Test, 29 Oct 2026) GPT-6 AstraWonder: 72.5 accuracy %, $5.6 per test (Vals.ai EMB v1.1 leaderboard, Cost / Test, 29 Oct 2026)Wonder

Excel Modeling Benchmark (EMB) v1.1

accuracy %, higher is better. 10 models.

Updated 29 Oct 2026

  1. Claude Fable 5.176.7
  2. Gemini 4 Carbon76.5
  3. Claude Sonnet 5.575.9
  4. Claude Opus 5.575.8
  5. Wonder Full75
  6. Gemini 4 Argon74.8
  7. GPT-6.1 Terra73.3
  8. Gemini 3.8 Flash72.9
  9. GPT-6 Astra72.7
  10. Wonder72.5

Excel Modeling Benchmark (EMB): cost

cost per test item (USD), lower is better. 10 models.

Updated 29 Oct 2026

  1. GPT-6.1 Terra0.560
  2. Wonder Full4.6
  3. Wonder5.6
  4. Gemini 4 Carbon5.8
  5. GPT-6 Astra5.8
  6. Gemini 3.8 Flash8.2
  7. Claude Sonnet 5.58.8
  8. Claude Opus 5.511.0
  9. Gemini 4 Argon11.8
  10. Claude Fable 5.115.9

Excel Modeling Benchmark (EMB) v1.1

accuracy %, higher is better. 10 models.

Updated 29 Oct 2026

  1. 1Claude Fable 5.176.7
  2. 2Gemini 4 Carbon76.5
  3. 3Claude Sonnet 5.575.9
  4. 4Claude Opus 5.575.8
  5. 5Wonder Full75
  6. 6Gemini 4 Argon74.8
  7. 7GPT-6.1 Terra73.3
  8. 8Gemini 3.8 Flash72.9
  9. 9GPT-6 Astra72.7
  10. 10Wonder72.5
Excel Modeling Benchmark (EMB) leaderboard (accuracy %, higher is better)
1Claude Fable 5.1Anthropic76.7v1.1 test setVals.ai leaderboard
Official board, 29 Oct 2026
2Gemini 4 CarbonGoogle76.5v1.1 test setVals.ai leaderboard
Official board, 29 Oct 2026
3Claude Sonnet 5.5Anthropic75.9v1.1 test setVals.ai leaderboard
Official board, 29 Oct 2026
4Claude Opus 5.5Anthropic75.8v1.1 test setVals.ai leaderboard
Official board, 29 Oct 2026
5Wonder FullNot stated75v1.1 test setVals.ai leaderboard
Official board, 29 Oct 2026
6Gemini 4 ArgonGoogle74.8v1.1 test setVals.ai leaderboard
Official board, 29 Oct 2026
7GPT-6.1 TerraOpenAI73.3v1.1 test setVals.ai leaderboard
Official board, 29 Oct 2026
8Gemini 3.8 FlashGoogle72.9v1.1 test setVals.ai leaderboard
Official board, 29 Oct 2026
9GPT-6 AstraOpenAI72.7v1.1 test setVals.ai leaderboard
Official board, 29 Oct 2026
10WonderNot stated72.5v1.1 test setVals.ai leaderboard
Official board, 29 Oct 2026

Last checked 1 Nov 2026. Something out of date or wrong? Send a correction.