Agentic Multimodal Tool-use on RealX-Bench benchmark leaderboard · SOTA2 Research

Agentic Multimodal Tool-use on RealX-Bench

70Average ScoreHuman

Evaluation Results

Method	Links
Human2025.11		70	69.5	63.5	62.1	51.4
Gemini 2.5 Pro2025.11		46	41.5	33.7	43.6	27.8
Gemini 2.5 Pro2025.11		45	37.8	33.2	43.1	25
Gemini 2.5 Pro2025.11		41.7	39	28.7	38.4	23.6
o32025.11		41.3	37.8	28.1	40.3	22.2
o32025.11		41	34.8	28.7	37.9	19.4
Gemini 2.5 Pro2025.11		39.3	34.8	24.2	36	16.7
o32025.11		39.3	38.4	25.3	37	20.8
GPT4o2025.11		38.7	30.5	27.5	36.5	15.3
GPT4o2025.11		36.3	29.3	25.8	36.5	16.7
o32025.11		35	31.7	23	30.8	11.1
GPT4o2025.11		32.3	29.9	22.5	29.4	16.7
GPT4o2025.11		32	29.3	23	29.4	16.7
Qwen2.5-VL-32B2025.11		32	27.4	29.2	31.8	23.6
Qwen2.5-VL-72B2025.11		31	35.4	25.8	25.6	23.6
Qwen2.5-VL-32B2025.11		30.7	27.4	23	26.1	15.3
DeepEyesV22025.11		28.3	19.5	22.5	28.9	18.1
Qwen2.5-VL-72B2025.11		28	28.7	20.2	20.4	15.3
Qwen2.5-VL-72B2025.11		26.3	28.7	19.7	20.4	16.7
Qwen2.5-VL-32B2025.11		25.7	25.6	20.2	19	16.7
Qwen2.5-VL-72B2025.11		25.3	23.1	17.4	17.5	9.7
Qwen2.5-VL-32B2025.11		25	21.3	19.7	19.9	12.5
Qwen2.5-VL-7B2025.11		22.3	17.1	16.3	19.9	9.7
Qwen2.5-VL-7B2025.11		21.7	17.7	15.7	18.5	7.6
Thyme2025.11		21	18.3	14.6	12.8	4.2
Qwen2.5-VL-7B2025.11		19.7	16.6	14.4	15.9	8.3
DeepEyes2025.11		19	19.5	14.6	12.8	9.7
Qwen2.5-VL-7B2025.11		17	15.9	13.5	12.3	6.9