Long-horizon Reasoning for Robotic Manipulation on RoboVQA benchmark leaderboard · SOTA2 Research

Long-horizon Reasoning for Robotic Manipulation on RoboVQA

75.1B-1 ScoreLoHo-Manip-4B

Evaluation Results

Method	Links
LoHo-Manip-4B2026.04		75.1	65	58.6	53.5	63.1
RynnBrain-8B2026.04		74.3	63.9	57.3	52.6	62.1
Qwen3-VL-8B2026.04		72.6	62.9	56.4	51.4	60.8
Fast-ThinkAct-3B2026.01		70.1	63	57.2	53	60.8
Fast-ThinkAct-3B2026.04		70.1	63	57.2	53	60.8
ThinkAct-7B2026.04		69.1	61.8	56	52.4	59.8
Qwen3-VL-4B2026.04		62.5	52.2	47	43.2	51.2
ThinkAct-3B2026.01		62.4	57.3	52	49.6	55.3
ThinkAct-3B2026.04		62.4	57.3	52	49.6	55.3
RoboBrain2.0-3B2026.01		54.4	47.7	43.1	41	46.5
RoboBrain2.0-3B2026.04		54.4	47.7	43.1	41	46.5
RoboBrain2.5-8B2026.04		48.8	38.2	33.2	29.8	37.5
Gemini-3.0-Flash2026.04		46.6	36.4	33.8	32.2	37.3
Qwen2.5-VL-3B2026.01		42.5	36.3	28.7	31.8	34.8
Qwen2.5-VL-3B2026.04		42.5	36.3	28.7	31.8	34.8
Gemini-2.5-Flash2026.01		39.1	31.6	22.9	22.1	28.9
Gemini-2.5-Flash2026.04		39.1	31.6	22.9	22.1	28.9
NVILA-2B2026.01		38.7	34.3	31.1	29.2	33.3
NVILA-2B2026.04		38.7	34.3	31.1	29.2	33.3
Magma-8B2026.01		38.6	31.5	28.1	26.7	31.2
Magma-8B2026.04		38.6	31.5	28.1	26.7	31.2
RoboBrain2.0-7B2026.04		37.4	31	27.1	25.8	30
InternVL2.5-2B2026.01		36.6	33.7	31	29.4	32.7
InternVL2.5-2B2026.04		36.6	33.7	31	29.4	32.7
InternVL3-2B2026.01		34.4	33.9	33.5	33.3	33.8
InternVL3-2B2026.04		34.4	33.9	33.5	33.3	33.8
GPT-4V2026.01		32.2	26.5	24.7	23.9	26.8
GPT-4V2026.04		32.2	26.5	24.7	23.9	26.8