Excel Modeling Benchmark v1.1 Scratch
Professional Work · 2026-10-19
Scratch-mode score on Excel Modeling Benchmark v1.1, where agents build the workbook without a template and are graded by the GPT-6 Luna partial-credit rubric.
Top models (higher is better)
| Model | Score |
|---|---|
| Claude Fable 5.1 | 72.7 |
| GPT-6 Astra | 70.9 |
| Claude Sonnet 5.5 | 70.8 |
| Claude Opus 5.5 | 70.2 |
| Muse Wonder Full | 70.0 |
| Gemini 4 Argon | 70.0 |
| Muse Wonder | 68.8 |
| Gemini 3.8 Flash | 68.5 |
| GPT-6 Luna | 68.1 |
| Grok 4.7 | 67.5 |
| DeepSeek V4.1 Flash 1017 | 67.3 |
| Hy4 Preview | 67.1 |
| GPT-6.1 Sol | 66.9 |
| Kimi K3 | 66.8 |
| Step 5 Preview | 66.6 |
| GLM-5.3 | 66.5 |
| Qwen3.8-Max | 66.3 |
| MiMo-V2.6-Flash | 65.7 |
| DeepSeek V4 Pro 0813 | 65.4 |
| DeepSeek V4.1 Flash | 65.2 |
| Mistral Large 4 Preview | 65.1 |
| MiniMax M3.1 Flash | 63.9 |
| MiMo-V2.6-Pro | 62.8 |
| Mercury 3 | 62.6 |
| GLM-5.3 Flash | 62.3 |
| Muse Spark Open | 61.7 |
| Beam 1.0 501B A23B | 61.0 |
| MiniMax M3 | 59.2 |
| Claude Haiku 5.5 | 57.7 |
| Inkling-Small | 43.5 |
| Inkling | 42.5 |