Reasoning on MME-RealWorld Lite benchmark leaderboard · SOTA2 Research

Reasoning on MME-RealWorld Lite

84OCR ScoreTTSP

Evaluation Results

Method	Links
TTSP2026.04		84	88	45.3	38.8	52.7
RTWI2026.04		83	84	42	36.5	50.1
PFlowNet2026.05		83	76	70	53.5	—
Self-Ref.2026.04		82	83	40	36.3	49.4
Deepconf2026.04		82	83	41.3	36.5	49.7
TTSP2026.04		81	83	44	34.3	49
SC2026.04		81	83	40.7	36	49.2
TRT2026.04		81	83	40.7	36	49.2
Self-Cer.2026.04		81	83	40.7	36.3	49.4
CISC2026.04		80	83	40.7	36	49.1
RTWI2026.04		79	80	43.3	32.8	47.2
SC2026.04		78	78	43.3	32.8	47
CISC2026.04		78	78	42.7	32.3	46.6
Self-Ref.2026.04		78	78	42.7	32.3	46.6
Self-Cer.2026.04		78	78	42.7	32.8	46.8
Deepconf2026.04		78	79	43.3	32.8	47.1
Thyme2026.05		78	71	48	36	—
TRT2026.04		77	79	43.3	32.8	47
InternVL3-38B2026.05		77	45	47.3	35	—
DeepEyes2026.04		76	69	44	35	46.8
LLaVA-OV-72B2026.05		76	41	38.7	39.3	—
InternVL3-78B2026.05		76	56	46	40.3	—
Qwen3-VL-32B2026.05		76	77	42.7	30	—
DeepEyes2026.05		76	69	44	35	—
Base2026.04		74	80	34	35.3	46.2
Qwen2.5-VL-32B2026.05		74	60	27.3	29.5	—
Qwen2.5-VL-72B2026.05		74	61	26.7	25.5	—
TreeVGR2026.05		74	66	51.3	39	—
Base2026.04		73	75	42.7	31	44.8
Qwen3-VL-8B2026.05		73	70	39.3	25.3	—
Qwen2.5-VL-7B2026.05		72	62	28.7	23	—
Qwen3-VL-4B2026.05		72	64	43.4	24.3	—
Pixel-Reasoner2026.04		71	72	46	32.5	45.6
Pixel-Reasoner2026.05		71	72	46	32.5	—
InternVL3-8B2026.05		70	44	40	37	—
DeepEyesV22026.05		70	76	44	37	—
LLaVA-OV-7B2026.05		65	33	38	32	—