Question Answering on TAT-QA (held-out) benchmark leaderboard · SOTA2 Research

Question Answering on TAT-QA (held-out)

59.33AccuracyClaude-3.5-Sonnet

Evaluation Results

Method	Links
Claude-3.5-Sonnet2025.09		59.33
Gemini-2.5-Pro2025.09		56.29
OpenAI-o4-mini2025.09		56.16
QVQ-72B-Preview2025.09		55.48
GPT-4o2025.09		53.85
Qwen2-VL-72B2025.09		52.23
Qwen2-VL-7B-GRPO2025.09		51.68
Qwen2-VL-7B-Table-R12025.09		48.06
Qwen2-VL-7B2025.09		46.5
Qwen2-VL-7B-SFT2025.09		37.82
Qwen2-VL-2B2025.09		30.44
Qwen2-VL-2B-GRPO2025.09		29.02
Qwen2-VL-2B-Table-R12025.09		26.42
DeepSeek-VL2 4.5B2025.09		24.11
Qwen2-VL-7B-Ins2025.09		20.85
Qwen2-VL-2B-SFT2025.09		19.69
Table-LLaVA 13B2025.09		15.67
Table-LLaVA 7B2025.09		12.82
Monkey 7B2025.09		12.31
LLaVA-v1.5 7B2025.09		2.97
mPLUG-Owl2 7B2025.09		0.39