FormationEval, an open multiple-choice benchmark for petroleum geoscience: leaderboard
FormationEval evaluates language models on multiple-choice petroleum geoscience and subsurface questions, using text input only. It covers seven domains such as petrophysics, petroleum geology, and reservoir engineering. The pinned metric is accuracy, measuring the rate of correct answers in the reported leaderboard results. The source notes public code and results, plus documented construction caveats. Source: BenchmarkList’s attributed result table.
Metric: accuracy (rate). Source: benchmarklist.com. Saturation forecast: Around November 2026. 72 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Gemini 3 Pro (Preview) | 1 | #71 |
| 2 | GLM-4.7 | 0.99 | #202 |
| 3 | Gemini 3 Flash (Preview) | 0.98 | #90 |
| 4 | Gemini 2.5 Pro | 0.98 | #159 |
| 5 | Grok 4.1 Fast | 0.98 | #227 |
| 6 | Kimi K2 (Thinking) | 0.97 | #252 (Kimi K2) |
| 7 | Claude Opus 4.5 | 0.97 | #87 |
| 8 | GPT-5 Mini (Medium) | 0.96 | #193 (GPT-5 Mini) |
| 9 | DeepSeek R1 | 0.96 | #262 |
| 10 | Grok 4 Fast | 0.96 | #266 |
| 11 | GPT-5 Mini (High) | 0.96 | #193 (GPT-5 Mini) |
| 12 | O4 Mini (High) | 0.95 | #185 (O4 Mini) |
| 13 | GPT-5 Mini (Low) | 0.95 | #193 (GPT-5 Mini) |
| 14 | Gemini 2.5 Flash | 0.95 | #255 |
| 15 | Grok 3 Mini | 0.95 | #289 |
No result here: #3 Claude Opus 5.5, #6 GPT-6 Astra, #7 GPT-6.1 Sol.
Interactive version: theaggregate.ai/benchmark?slug=formationeval-an-open-multiple-choice-benchmark-for-petroleum-geoscience · How It Works · Data refreshed daily, snapshot 2026-11-05.