Multimodal Comment Generation on HotComment (test) benchmark leaderboard · SOTA2 Research

Multimodal Comment Generation on HotComment (test)

19.27BLEU-1Qwen3-VL-8B + StyleCmt

Evaluation Results

Method	Links
Qwen3-VL-8B + StyleCmt2026.04		19.27	19.53	62.78	60.98	82.06	84.09
LLaMA3.2-Vision + StyleCmt2026.04		17.58	16.89	62.57	57.42	75.07	75.04
Mistral-3.1-24B2026.04		15.32	16.07	60.68	54.25	74.22	76.08
Gemini2.5-Image2026.04		15.25	13.69	61.34	56.63	72.05	73.09
ChatGPT-4o2026.04		14.84	12.65	61.21	55.71	72.06	73.01
Qwen3-VL-8B2026.04		14.55	15.59	59.03	49.81	72.52	70.58
LLaMA3.2-Vision2026.04		13.82	13.84	60.04	49.43	62.03	65.08
ChatGPT-4o-mini2026.04		13.25	10.47	61.21	52.56	70.03	71.07
Qwen3-VL-4B2026.04		12.75	13.31	58.42	34.38	68.03	64.22