Counter-Perception Discrimination on CP-Bench (test) benchmark leaderboard · SOTA2 Research

Counter-Perception Discrimination on CP-Bench (test)

86F1 ScoreClaude-3-5-Sonnet

Evaluation Results

Method
Claude-3-5-Sonnet2025.04	86	85.3	82.4	90
GPT-4o-08062025.04	85.5	85	82.5	88.8
LLaVA-v1.5-13B + Antidote2025.04	83.5	84.5	89.5	78.3
GLM-4v2025.04	79.7	80.3	82.1	79.7
LLaVA-v1.5-7B + Antidote2025.04	78.4	84.5	73.1	73.1
GPT-4o-mini-07182025.04	77.6	75.9	72.6	83.3
LLaVA-Next-Mistral-7B + Antidote2025.04	76.8	77.5	79.4	74.3
GPT-4v-04092025.04	71.6	75.8	86.6	61
InternVL-2-Pro2025.04	64.2	70.5	81.5	53
LLaVA-Next-Vicuna-13B2025.04	56	67.3	85.5	41.6
MiniCPM-V2.5-8B2025.04	55.2	66.1	81.3	41.8
Qwen2-VL-72B2025.04	48.8	65	90.7	33.3
LLaVA-Next-Vicuna-7B2025.04	47	63.6	86.1	32.3
InternVL2-26B2025.04	38.5	59.6	80.8	25.3
InstructBLIP-7B2025.04	35.6	47.4	45.9	29.1
InternVL2-8B2025.04	34.3	58.4	81.8	21.7
Cogvlm2-19B2025.04	34.2	57.4	75.3	22.1
LLaVA-Next-Mistral-7B2025.04	26.7	54.8	70.7	16.5
Phi-3-Vision2025.04	26.1	54	66.4	16.3
LLaVA-v1.5-7B + SeVa2025.04	24.1	55.1	78	14.3
Qwen2-VL-7B2025.04	22.7	54.9	80.3	13.2
LLaVA-v1.5-13B2025.04	17.3	53.8	82.8	9.6
LLaVA-v1.5-7B + VCD2025.04	6.1	50.7	64	3.2
LLaVA-v1.5-7B2025.04	5.7	50.5	60	3
LLaVA-v1.5-7B + HA-DPO2025.04	4.7	50.6	66.7	2.4