Stratigraphic event reconstruction on geo-strat-rl diagram images (test) benchmark leaderboard · SOTA2 Research

Stratigraphic event reconstruction on geo-strat-rl diagram images (test)

0.747R (Structural Fidelity)GPT-5.5 (OpenAI, 2026)

Evaluation Results

Method	Links
GPT-5.5 (OpenAI, 2026)2026.06		0.747	0.668	1	0.917	0.681	0.867	0.756	0.435	0.627	0.391	0.991	1	0	0
Qwen3-VL-4B2026.06		0.74	0.664	1	0.906	0.633	0.981	0.755	0.428	0.618	0.328	0.964	1	0	0
Gemma 4 31B IT2026.06		0.701	0.616	0.992	0.867	0.603	0.836	0.72	0.403	0.608	0.275	0.986	0.992	0	0
Qwen2.5-VL-3B2026.06		0.695	0.609	1	0.812	0.603	0.861	0.75	0.408	0.601	0.229	0.905	0.75	0	0
Gemini 2.5 Flash Lite2026.06		0.691	0.613	0.978	0.78	0.62	0.818	0.731	0.423	0.615	0.307	0.834	0.978	0	0.022
Mistral Small 3.2 24B2026.06		0.624	0.563	0.966	0.483	0.526	0.94	0.74	0.403	0.588	0.264	0.659	0.725	0.014	0.02
Qwen3-VL-30B-A3B2026.06		0.621	0.552	0.872	0.706	0.525	0.817	0.665	0.384	0.548	0.218	0.768	0.872	0	0.02
Qwen3-VL-4B2026.06		0.311	0.287	0.43	0.355	0.256	0.377	0.309	0.233	0.29	0.191	0.424	0.376	0	0
Qwen2.5-VL-3B2026.06		0.208	0.189	0.404	0.232	0.121	0.314	0.139	0.167	0.246	0.107	0.332	0.328	0	0