Analytical QA on MuDABench Simple split benchmark leaderboard · SOTA2 Research

Analytical QA on MuDABench Simple split

89.4Process AccuracyHuman Performance

Evaluation Results

Method	Links
Human Performance2026.04		89.4	83.34	73.34
Noise WF w/ GPT 4.1 mini + Chunk = 52026.04		59.61	16.36	7.27
WF w/ GPT 4.1 mini + Chunk = 52026.04		58.88	22.43	7.48
WF w/ GPT 4.1 mini + Chunk = 32026.04		58.03	24.3	6.54
WF w/ GPT 4o + Chunk = 12026.04		41.79	6.67	0
GPT 4o + Chunk = 2.5 \|D\| + Metadata2026.04		25.14	13.25	5.42
GPT 4o + Chunk = 2.5 \|D\|2026.04		21.63	10.84	3.01
GPT 4o + Chunk = 2 \|D\| + Metadata2026.04		19.78	13.86	4.22
GPT 4o + Chunk = 2 \|D\|2026.04		17.93	12.65	4.22
GPT 4o + Chunk = 1.5 \|D\|2026.04		17.61	9.64	3.01
GPT 4o + Chunk = 1.5 \|D\| + Metadata2026.04		16.2	11.45	3.01
GPT 4o + Chunk = 1 \|D\|2026.04		15.72	6.63	2.41
GPT 4o + Chunk = 1 \|D\| + Metadata2026.04		13.38	10.84	4.22