Astronomical Spectrum Classification on DESI (unseen)
82.6CV PerformanceSpec-o3-7B
Evaluation results
| Method | Links | ||||
|---|---|---|---|---|---|
| Spec-o3-7BParameters=7B2026.01 | 82.6 | 72.8 | 76.7 | 77.4 | |
| Spec-o3-3BParameters=3B2026.01 | 76.1 | 70.1 | 74.5 | 73.6 | |
| o3Model Type=Proprietary VLM2026.01 | 60.2 | 51.4 | 56.8 | 56.1 | |
| CarbonNetModel Type=Specialist Deep Learning Model2026.01 | 49.6 | 64.1 | 58.2 | 57.3 | |
| AstroCLIPModel Type=Specialist Deep Learning Model2026.01 | 48.3 | 53.9 | 50.8 | 51 | |
| GaiaNetModel Type=Specialist Deep Learning Model2026.01 | 48.3 | 53.8 | 43.7 | 48.6 | |
| Qwen3-VL-8BModel Type=Open-source VLM, Parameters=8B2026.01 | 33.8 | 23.9 | 30.4 | 29.4 | |
| Qwen3-VL-30B-A3BModel Type=Open-source VLM, Parameters=30B2026.01 | 31.5 | 31.1 | 30.2 | 30.9 | |
| GPT-4.1Model Type=Proprietary VLM2026.01 | 28.9 | 34.4 | 31.8 | 31.7 | |
| Qwen2.5-VL-7BModel Type=Open-source VLM, Parameters=7B2026.01 | 25.9 | 33.7 | 30.6 | 30.1 | |
| Qwen2.5-VL-3BModel Type=Open-source VLM, Parameters=3B2026.01 | 23.3 | 30.8 | 23.1 | 25.7 |