Question Answering on TabMWP (held-in) benchmark leaderboard · SOTA2 Research

Question Answering on TabMWP (held-in)

92.6AccuracyQwen2-VL-7B-Table-R1

Evaluation Results

Method	Links
Qwen2-VL-7B-Table-R12025.09		92.6
Qwen2-VL-7B-SFT2025.09		90.3
Gemini-2.5-Pro2025.09		89.9
Qwen2-VL-7B-GRPO2025.09		89.2
GPT-4o2025.09		87.59
OpenAI-o4-mini2025.09		86.7
QVQ-72B-Preview2025.09		86.2
Claude-3.5-Sonnet2025.09		83.3
Qwen2-VL-2B-Table-R12025.09		83.2
Qwen2-VL-72B2025.09		81.95
Qwen2-VL-2B-GRPO2025.09		71.4
Qwen2-VL-2B-SFT2025.09		70
Qwen2-VL-7B2025.09		63.8
Table-LLaVA 13B2025.09		59.77
Table-LLaVA 7B2025.09		57.78
DeepSeek-VL2 4.5B2025.09		53.75
Qwen2-VL-7B-Ins2025.09		49.51
Qwen2-VL-2B2025.09		46.1
Monkey 7B2025.09		13.26
mPLUG-Owl2 7B2025.09		6.83
LLaVA-v1.5 7B2025.09		6.05