List your company

Human Preference Alignment on MM-AlignBench 1.0 (test)

84.9Win RateClaude3.5V-Sonnet
-0.58821.60643.865.994Feb 25, 2025

Evaluation results

MethodLinks
Claude3.5V-Sonnet2025.0284.951.47014412314
GPT-4V2025.0282.5465715712311
GPT-4o2025.0281.3498112412314
GeminiFlash1.5-0022025.027739.15613814359
LLaVANext-OA-32B-DPOAlignment Stage=SFT+DPO, Dataset=OmniAlign-V + OmniAlign-V-DPO2025.0272.633.54913820405
LLaVANext-OA-32BAlignment Stage=SFT, Dataset=OmniAlign-V2025.0262.319.431126196214
Qwen2VL-72B2025.0261.521.64311215757
Claude-3V-Sonnet2025.02500—————
Qwen2VL-7B2025.0244.4-5.82884510134
InternVL2-72B2025.0244.4-6.9199389834
InternVL2-8B-MPO2025.0240.1-10.926751010041
InternVL2-8B2025.0231.3-21.818611510949
LLaMA3.2-Vision-11B2025.0227.8-33.7185249880
LLaVANext-Qwen32B2025.0226.6-2916511012154
LLaVA-OneVision-7B2025.0223.8-46.21446175116
MiniCPM-V-2.52025.0212.7-53923811696
Xcomposer2.5-7B2025.027.5-74514363167
Idefics3-8B2025.022.7-92.334015230