FinSheet-Bench: leaderboard
Metric: Accuracy (%) over all 501 questions (16 templates from simple lookups to complex aggregation) on the 24 synthetic private-equity fund portfolio spreadsheets, each serialized into the prompt zero-shot; a question on a file that exceeds the model's context window counts as wrong (effective accuracy); answers checked by exact matching, fuzzy matching and GPT-4o-mini/Gemini 3 Flash adjudication; higher is better. Source: arxiv.org. Saturation forecast: Around January 2027. 10 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Gemini 3.1 Pro (Preview) | 82.4 | #54 |
| 2 | GPT-5.2 (Thinking) | 80.4 | #105 (GPT-5.2) |
| 3 | Gemini 3 Pro | 80.2 | #77 |
| 4 | Claude Opus 4.6 (Thinking) | 80.2 | #60 (Claude Opus 4.6) |
| 5 | Gemini 2.5 Pro | 78.8 | #145 |
| 6 | Claude Opus 4.6 | 66.7 | #60 |
| 7 | GPT-5.2 (Non-reasoning) | 57.7 | #105 (GPT-5.2) |
| 8 | GPT-4o | 54.1 | #333 |
| 9 | Gemini 2.0 Flash | 50.1 | #331 |
| 10 | GPT-3.5 Turbo | 24.8 | #849 |
No result here: #3 Claude Opus 5.5, #5 GPT-6 Astra, #8 Claude Fable 5.1.
Interactive version: theaggregate.ai/benchmark?slug=finsheet-bench · How It Works · Data refreshed daily, snapshot 2026-10-11.