Long-horizon Ego-Centric Spatiotemporal Reasoning on Dual-Level Evaluation Framework Long level benchmark leaderboard · SOTA2 Research

Long-horizon Ego-Centric Spatiotemporal Reasoning on Dual-Level Evaluation Framework Long level

98.8Forward Accuracy (S)INTERNVL-8B

Evaluation Results

Method	Links
INTERNVL-8B2026.04		98.8	99.5	99.7	1.5	2.2	2.2	50.6
HUMAN PERCEPTION2026.04		94.3	95.5	95.6	93.9	96.1	95.9	95.2
QWEN2.5-VL-7B2026.04		93.1	89.1	85	8.3	11.9	13.6	49.8
EGOTSR-COT2026.04		91.9	90.1	93	1.8	2.2	2.6	46.9
EGOTSR-LONGTAG2026.04		88.2	92.9	96.2	92	92.7	92.3	92.4
3D-LLM2026.04		86.8	90.5	84.3	7.5	9.6	7.6	47.7
NVILA-8B2026.04		77.8	80.5	78.4	21.9	19	16	49.8
PALIGEMMA2026.04		76	81.9	87.6	16.8	14.1	16	49.5
DEEPSEEK-VL22026.04		76	82.4	87.6	17.1	14.1	16	49.7
GEMINI-2.5-PRO2026.04		74.1	78.1	64	63.2	42.1	57.5	62
LLAVA-ONEVIS2026.04		73.7	80	86.4	19.2	13.8	18.5	49.3
GPT-O32026.04		71.9	64.7	73.1	69.6	66.7	68.6	69
DOUBAO-1.5-PRO2026.04		68.4	58.3	58.8	45.2	48.2	29.4	52
GPT-O4MINI2026.04		64.9	69	85.7	34.6	30.3	42.5	55
GPT-4O2026.04		56.6	61.2	69.7	57.3	51.5	54.6	58.2
EGOTSR-TAG2026.04		55	49	58.7	62.3	54.9	54.9	55.8
SEED-1.62026.04		46.5	46.3	28.6	65.4	59.4	52	49.2
VIDEO-3D LLM2026.04		25.2	22.5	20.8	74.6	79.9	78.3	50.2
CHAT-SCENE2026.04		24.3	18.1	12.4	82.9	85.3	83.4	50.3
QWEN2.5-VL-3B2026.04		24	17.3	12.1	82.9	84.4	83.1	49.9
LL3DA2026.04		24	17.6	12.7	83.2	85.6	83.7	50.4
3D-LLAVA2026.04		11.3	11.8	12.4	86.2	86.5	85.9	48.1