Instruction Generation on Eval-400 In-house (test)
66CorrectnessGemini-3-Pro
Evaluation results
| Method | Links | ||||
|---|---|---|---|---|---|
| Gemini-3-Pro2026.04 | 66 | 21 | 12 | 1 | |
| Qwen3-VL-235B-A22B (SFT+DPO)Training=SFT+DPO2026.04 | 66 | 23 | 10.4 | 0.6 | |
| Qwen3-VL-32B (SFT+DPO)Training=SFT+DPO2026.04 | 57 | 29 | 13.25 | 0.75 | |
| GPT-4.12026.04 | 48.75 | 42.25 | 8.25 | 0.75 | |
| Qwen3-VL-235B-A22BMode=base2026.04 | 41.75 | 47.75 | 8.5 | 2 | |
| GLM-4.5V (106B-A12B)2026.04 | 38.6 | 51.2 | 9.2 | 1 | |
| Qwen3-VL-32BMode=base2026.04 | 37 | 51.5 | 11 | 0.5 |