Excel Modeling Benchmark Scratch
Professional Work · 2026-07-01
Scratch-mode score on Excel Modeling Benchmark, where agents build a financial model without a template workbook and are judged against the task rubric.
Top models (higher is better)
| Model | Score |
|---|---|
| Claude Opus 5 | 72.3 |
| Claude Fable 5 | 70.0 |
| GPT-5.6 Sol | 69.1 |
| GPT-5.6 Luna | 68.8 |
| Kimi K3 | 68.2 |
| Opus 4.8 | 67.4 |
| Grok 4.5 | 67.3 |
| GPT-5.5 | 66.8 |
| GLM-5.2 | 66.7 |
| Gemini 3.6 Flash | 65.7 |
| Sonnet 5 | 65.0 |
| Sonnet 4.6 | 64.8 |
| GPT-5.6 Terra | 64.7 |
| Gemini 3.5 Flash | 64.0 |
| Muse Spark 1.1 | 63.3 |
| Qwen3.7-Max | 61.4 |
| MiniMax M3 | 61.0 |
| Kimi K2.6 | 59.7 |
| MiMo-V2.5-Pro | 57.9 |
| Gemini 3.1 Pro Preview | 57.7 |
| Qwen3.7-Plus | 57.0 |
| DeepSeek-V4-Pro | 56.7 |
| GPT-5.4 Mini | 54.7 |
| Inkling-Small | 45.1 |
| Gemini 3.5 Flash-Lite | 44.9 |
| GPT-5.4 Nano | 43.8 |
| Inkling | 43.0 |
| Grok 4.3 | 29.4 |
| Gemini 3.1 Flash-Lite Preview | 9.1 |