Stratigraphic Event Reconstruction on geo-strat-rl seismic-style observations (test)
0.745R ScoreQwen3-VL-4B
Evaluation results
| Method | Links | ||||||||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Qwen3-VL-4BClass=open VLM, Method=seismic ng16 LoRA2026.06 | 0.745 | 0.67 | 1 | 0.863 | 0.663 | 0.974 | 0.768 | 0.457 | 0.603 | 0.36 | 0.935 | 1 | 0 | 0 | |
| Mistral Small 3.2 24BClass=open VLM, Method=pretrained2026.06 | 0.716 | 0.636 | 1 | 0.823 | 0.64 | 0.871 | 0.775 | 0.434 | 0.6 | 0.308 | 0.9 | 0.75 | 0 | 0 | |
| GPT-5.5 (OpenAI, 2026)Class=closed VLM, Method=medium reasoning2026.06 | 0.656 | 0.59 | 1 | 0.732 | 0.552 | 0.773 | 0.604 | 0.459 | 0.637 | 0.372 | 0.944 | 1 | 0 | 0 | |
| Qwen3-VL-30B-A3BClass=open VLM, Method=pretrained2026.06 | 0.655 | 0.582 | 0.928 | 0.761 | 0.55 | 0.853 | 0.679 | 0.398 | 0.58 | 0.254 | 0.868 | 0.928 | 0 | 0.022 | |
| Qwen2.5-VL-3BClass=open VLM, Method=diagram ng16 LoRA2026.06 | 0.655 | 0.588 | 0.998 | 0.758 | 0.492 | 0.905 | 0.66 | 0.428 | 0.6 | 0.275 | 0.964 | 0.748 | 0 | 0 | |
| Gemma 4 31B ITClass=open VLM, Method=pretrained2026.06 | 0.627 | 0.545 | 0.994 | 0.71 | 0.483 | 0.71 | 0.675 | 0.42 | 0.603 | 0.215 | 0.992 | 0.994 | 0 | 0 |