Game-based visual-agent task completion on VAB-Minecraft
73.28Success RateGemini 3 Flash
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| Gemini 3 FlashSkill condition=MMSkills2026.05 | 73.28 | 0.7884 | |
| Gemini 3 FlashSkill condition=Text-only2026.05 | 68.96 | 0.7541 | |
| GLM-5VSkill condition=MMSkills2026.05 | 68.1 | 0.7495 | |
| Gemini 3 FlashSkill condition=No skill2026.05 | 67.24 | 0.7462 | |
| Qwen3-VL-235BSkill condition=MMSkills2026.05 | 62.07 | 0.7114 | |
| GLM-5VSkill condition=Text-only2026.05 | 61.2 | 0.6938 | |
| GLM-5VSkill condition=No skill2026.05 | 56.03 | 0.6701 | |
| Qwen3-VL-235BSkill condition=Text-only2026.05 | 55.17 | 0.6634 | |
| Qwen3-VL-235BSkill condition=No skill2026.05 | 52.59 | 0.6308 | |
| Qwen3-VL-8B-InstructSkill condition=MMSkills2026.05 | 38.79 | 0.4668 | |
| Qwen3-VL-8B-InstructSkill condition=Text-only2026.05 | 29.31 | 0.3754 | |
| Qwen3-VL-8B-InstructSkill condition=No skill2026.05 | 23.28 | 0.3017 |