FormationEval
Science · 2026-01-05
FormationEval is a 505-question multiple-choice benchmark for petroleum geoscience and adjacent subsurface disciplines. Its evaluated v0.1 track spans seven domains, including petrophysics, petroleum geology, geophysics, and reservoir engineering.
Top models (higher is better)
| Model | Score |
|---|---|
| Gemini 3 Pro Preview | 99.8 |
| GLM-4.7 | 98.6 |
| Gemini 3 Flash Preview | 98.2 |
| Gemini 2.5 Pro | 97.8 |
| Grok 4.1 Fast | 97.6 |
| GPT-5.2 | 97.4 |
| Kimi K2 Thinking | 97.2 |
| Opus 4.5 | 97.0 |
| GPT-5 Mini | 96.4 |
| GPT-5.1 | 96.4 |
| DeepSeek-R1 | 96.2 |
| Grok 4 Fast | 96.0 |
| O4 Mini | 95.2 |
| Gemini 2.5 Flash | 95.0 |
| Grok 3 Mini | 95.0 |
| DeepSeek-V3.2 | 94.9 |
| o3-mini | 94.9 |
| Claude 3.7 Sonnet | 94.7 |
| GPT-5 Chat (2025-08-07) | 94.5 |
| GPT-4.1 | 93.7 |
| Gemini 2.0 Flash 001 | 93.3 |
| GPT-5 Nano | 93.3 |
| Llama 4 Scout Instruct | 93.1 |
| Mistral Medium 3.1 | 93.1 |
| Qwen3 235B A22B Instruct 2507 | 93.1 |
| Qwen3 30B A3B Thinking 2507 | 93.1 |
| GPT-4o | 92.9 |
| MiniMax M2 | 92.9 |
| Qwen3 14B | 92.9 |
| Qwen3 32B | 92.1 |
| GPT-4.1 Mini | 91.7 |
| Haiku 4.5 | 91.5 |
| Gemini 2.5 Flash-Lite | 91.3 |
| gpt-oss-120b | 90.7 |
| Qwen3 VL 8B Thinking | 90.3 |
| Mistral Small 3.2 24B Instruct 2506 | 89.3 |
| gpt-oss-20b | 89.3 |
| Sonnet 4.5 | 89.1 |
| Mistral Small 3 24B Instruct 2501 | 88.7 |
| Qwen3 8B | 88.7 |