Atlas

Benchmarks

← All benchmarks

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)

ModelScore
Gemini 3 Pro Preview99.8
GLM-4.798.6
Gemini 3 Flash Preview98.2
Gemini 2.5 Pro97.8
Grok 4.1 Fast97.6
GPT-5.297.4
Kimi K2 Thinking97.2
Opus 4.597.0
GPT-5 Mini96.4
GPT-5.196.4
DeepSeek-R196.2
Grok 4 Fast96.0
O4 Mini95.2
Gemini 2.5 Flash95.0
Grok 3 Mini95.0
DeepSeek-V3.294.9
o3-mini94.9
Claude 3.7 Sonnet94.7
GPT-5 Chat (2025-08-07)94.5
GPT-4.193.7
Gemini 2.0 Flash 00193.3
GPT-5 Nano93.3
Llama 4 Scout Instruct93.1
Mistral Medium 3.193.1
Qwen3 235B A22B Instruct 250793.1
Qwen3 30B A3B Thinking 250793.1
GPT-4o92.9
MiniMax M292.9
Qwen3 14B92.9
Qwen3 32B92.1
GPT-4.1 Mini91.7
Haiku 4.591.5
Gemini 2.5 Flash-Lite91.3
gpt-oss-120b90.7
Qwen3 VL 8B Thinking90.3
Mistral Small 3.2 24B Instruct 250689.3
gpt-oss-20b89.3
Sonnet 4.589.1
Mistral Small 3 24B Instruct 250188.7
Qwen3 8B88.7
Loading Atlas data…