Multi-turn conversation on ConvBench 1.0 (test) benchmark leaderboard · SOTA2 Research

Multi-turn conversation on ConvBench 1.0 (test)

39.51R1 (Pairwise)GPT-4V

Evaluation Results

Method	Links
GPT-4V2024.04		39.51	38.47	38.47	39.34	37.61	40.55	7.09	7.3	7.3	7.48	7.12	6.88
Claude-3 Opus2024.04		36.6	37.49	38.99	39.17	34.32	35.7	6.54	6.75	6.53	7.04	6.68	6.32
Reka Flash2024.04		25.6	24.67	25.13	27.56	21.32	26.52	6.78	6.86	6.93	7.25	6.41	6.7
InternVL 1.22024.04		21.17	22.41	24.96	21.31	20.97	19.93	5.49	5.69	5.8	5.88	5.39	5.29
InternVL 1.52024.04		17.65	20.22	26	17.33	17.33	15.08	5.6	5.76	6.11	5.93	5.25	5.43
ShareGPT4V-13B2024.04		17.56	17.45	17.85	18.72	15.77	17.68	4.85	5.03	5.16	5.06	4.86	4.67
LLaVA-1.5-13B2024.04		16.93	18.08	20.45	18.02	15.77	15.77	4.94	5.14	5.03	5.41	4.99	4.74
XComposer22024.04		15.83	16.41	17.16	19.06	13	15.25	5.82	5.98	5.98	6.17	5.78	5.66
mPLUG-Owl22024.04		14.93	15.83	17.5	17.16	12.82	14.04	5.04	5.17	4.98	5.38	5.14	4.91
Qwen-VL-Chat2024.04		14.33	14.62	16.29	18.37	9.19	14.04	5.54	5.65	5.96	5.22	—	5.43
MiniGPT-42024.04		10.95	10.8	11.61	11.27	9.53	11.09	3.85	4.04	3.99	4.4	3.73	3.66
LLaMA-A-V22024.04		9.04	9.59	8.84	10.92	9.01	8.49	4.77	4.91	4.77	5.47	4.48	4.64
Gemini Pro 1.02024.04		8.44	8.55	9.01	9.36	7.28	8.32	4.42	4.6	5.18	4.95	3.66	4.24