Reward Modeling on FIRM-Bench Edit
0.53Execution MAEFIRM-Edit-8B
Evaluation results
| Method | Links | |||
|---|---|---|---|---|
| FIRM-Edit-8B2026.03 | 0.53 | 0.73 | 0.62 | |
| Gemini-3-pro2026.03 | 0.54 | 0.57 | 0.55 | |
| GPT-52026.03 | 0.62 | 0.73 | 0.67 | |
| Qwen3-VL-8B2026.03 | 0.66 | 1.12 | 0.87 | |
| InternVL3.5-241B-A282026.03 | 0.69 | 1.05 | 0.86 | |
| Qwen3-VL-32B2026.03 | 0.69 | 1.14 | 0.9 | |
| Qwen3-VL-235B-A222026.03 | 0.72 | 0.91 | 0.81 | |
| GPT-4o2026.03 | 0.73 | 1.16 | 0.93 | |
| GPT-4.12026.03 | 0.74 | 1.04 | 0.88 | |
| InternVL3.5-38B2026.03 | 0.85 | 1.22 | 1.02 | |
| InternVL3.5-8B2026.03 | 0.9 | 1.13 | 1 |