End-to-End Extraction on marine Pb benchmark
50.8PrecisionCompass
Evaluation Results
| Method | Links | |||
|---|---|---|---|---|
| CompassSize=32B2026.05 | 50.8 | 42.9 | 46.5 | |
| GPT-4oSize=N/A2026.05 | 42.9 | 33 | 37.3 | |
| Gemini-2.5-proSize=N/A2026.05 | 33.4 | 51.1 | 40.4 | |
| Qwen2.5Size=32B2026.05 | 32 | 39.4 | 35.3 | |
| CompassSize=8B2026.05 | 19.6 | 15.8 | 17.5 | |
| Qwen3Size=8B2026.05 | 17.5 | 22.3 | 19.6 | |
| Llama-3.1Size=8B2026.05 | 6.7 | 8.5 | 7.5 | |
| OceanGPTSize=7B2026.05 | 3.8 | 0.5 | 0.9 | |
| K2 (LLaMA)Size=7B2026.05 | 0 | 0 | 0 |