List your company

Document Question Answering on DUDE (test)

65.931ANLSInternVL2-8B-CoB
13.3475626.9990340.650554.30197May 24, 2025Jul 17, 2025Sep 9, 2025Nov 2, 2025Dec 26, 2025Feb 18, 2026Apr 14, 2026

Evaluation results

MethodLinks
InternVL2-8B-CoBDoc-CoB=true2025.0565.931
InternVL2-40B2025.0565.802
GPT-4o2025.0565.57
Qwen2.5-VL-CoBDoc-CoB=true2025.0564.36
InternVL2-8B*SFT baseline=true2025.0564.12
Qwen2.5-VL*SFT baseline=true2025.0563.88
InternVL2-8B-RReasoning tasks=true2025.0563.42
Qwen2.5-VL-RReasoning tasks=true2025.0562.73
InternVL2-2B-CoBDoc-CoB=true2025.0559.92
InternVL2-2B*SFT baseline=true2025.0558.84
InternVL2-2B-RReasoning tasks=true2025.0558.31
DocOwl 1.5-CoBDoc-CoB=true2025.0557.87
DocOwl 1.5-RReasoning tasks=true2025.0557.52
DocOwl 1.5*SFT baseline=true2025.0557.21
Visual CoT2025.0553.41
TextCoT2025.0543.68
MultiDocFusionmax_len=550 tokens2026.0417.93
Structure based Chunkingmax_len=550 tokens2026.0416.83
Perplexity chunkingmax_len=550 tokens2026.0416.68
Length chunkingmax_len=550 tokens2026.0415.92
LumberChunkermax_len=550 tokens2026.0415.73
Semantic chunkingmax_len=550 tokens2026.0415.37