Spreadsheet Arena: leaderboard
Models generate complete spreadsheet workbooks from user prompts across finance, operations, business, research and creative use cases. Human voters compare rendered outputs in 555 public battles spanning 124 prompts. Score is the fraction of each model appearances earning outright preference; ties and both-bad remain denominator but not wins. This descriptive fraction differs from the paper fitted Bradley-Terry score. Existing minimum 10 battles policy is retained. Exact evaluated API checkpoint is used.
Metric: Outright preference wins / all battle appearances (%). Source: huggingface.co. Saturation forecast: Around January 2027. 14 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Claude Opus 4.5 (20251101) | 73.56 | #74 |
| 2 | Claude Sonnet 4.5 | 69.32 | #142 |
| 3 | Claude Opus 4.1 (20250805) | 64.29 | #132 |
| 4 | GPT-5.2 | 60.81 | #110 |
| 5 | Gemini 3 Pro (Preview) | 54.65 | #66 |
| 6 | Grok 4.1 Fast (Reasoning) | 49.23 | #217 (Grok 4.1 Fast) |
| 7 | GPT-5 | 37.5 | #98 |
| 8 | Gemini 2.5 Pro | 35.71 | #152 |
| 9 | GPT-5.1 (2025-11-13) | 33.98 | #101 |
| 10 | Grok Code Fast 1 | 25.77 | #337 |
| 11 | Kimi K2 0905 | 16.67 | #288 |
| 12 | GPT-4o (2024-08-06) | 11.58 | #336 |
| 13 | Qwen 3 30B A3B | 3.28 | #493 |
| 14 | Llama 4 Maverick Instruct | 3.17 | #456 |
No result here: #3 Claude Opus 5.5, #6 GPT-6 Astra, #7 GPT-6.1 Sol.
Interactive version: theaggregate.ai/benchmark?slug=spreadsheet-arena · How It Works · Data refreshed daily, snapshot 2026-10-20.