Beyond Simply Environment Scaling: Designing Effective Environment Distributions for Multimodal Agent Learning

Open paper
Benchmark Model Rank Results
multimodal-reasoning-on-mathvisionAES+HDC (Qwen3-VL-8B-Instruct)#27Accuracy: 54.9
multimodal-reasoning-on-mmmuAES+HDC (Qwen3-VL-8B-Instruct)#35Accuracy: 70.1
vision-language-understanding-on-mmstarAES+HDC (Qwen3-VL-8B-Instruct)#71Accuracy: 63.5

Rank counts only results with a code link.