OenoBench (Expert): leaderboard
Wine expertise built from a 38,000-fact corpus scraped out of appellation registries, consortia and academic sources: regions, grape varieties, producers, viticulture, winemaking and the wine business. This board is the expert tier (L4) only — the entry tier saturates above 95% and the paper publishes no overall column, so the hard tier is where the models separate.
Metric: Accuracy (%, L4 expert tier). Source: arxiv.org. Saturation forecast: Around February 2027. 16 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | O3 | 71.2 | #124 |
| 2 | Gemini 2.5 Pro (Thinking) | 69.4 | #152 (Gemini 2.5 Pro) |
| 3 | GPT-5 | 69.2 | #98 |
| 4 | Claude Opus 4.7 | 69.1 | #49 |
| 5 | Claude Opus 4.7 (Thinking) | 68.6 | #49 (Claude Opus 4.7) |
| 6 | Gemini 2.5 Pro | 68.5 | #152 |
| 7 | GPT-5 Mini | 64.8 | #186 |
| 8 | DeepSeek R1 | 62.1 | #251 |
| 9 | Gemini 2.5 Flash | 60.1 | #245 |
| 10 | Mistral Large 2 (Nov) Instruct (2411) | 54 | #465 |
| 11 | DeepSeek V3 | 52.4 | #321 |
| 12 | Qwen 2.5 72B | 51.4 | #499 |
| 13 | Llama 3.3 70B | 50.3 | #579 |
| 14 | Llama 3.1 8B | 45 | #1151 |
| 15 | Qwen 2.5 7B | 43.3 | #922 |
No result here: #3 Claude Opus 5.5, #6 GPT-6 Astra, #7 GPT-6.1 Sol.
Interactive version: theaggregate.ai/benchmark?slug=oenobench-expert · How It Works · Data refreshed daily, snapshot 2026-10-20.