Share your thoughts, 1 month free Claude Pro on usSee more

Spreadsheet Understanding and Manipulation on SpreadsheetBench Hard Restriction

66.67Cell-Level Accuracy

Human Performance

Updated 4d ago

Evaluation Results

Method	Links
Human Performance 2026.04		66.67	55	62
SpreadsheetAgent 2026.04		36.9	31.05	34.65
SpreadsheetAgent 2026.04		32.8	29.34	31.47
TreeThinker 2026.04		31.91	27.92	30.37
SpreadsheetAgent 2026.04		27.81	25.64	29.25
SpreadsheetAgent 2026.04		25.31	25.07	25.22
GPT-OSS-120B 2026.04		24.96	23.93	24.56
TreeThinker 2026.04		24.42	26.5	25.22
Qwen3-Coder-480B 2026.04		22.82	27.07	24.45
GPT-OSS-20B 2026.04		19.43	21.37	20.18
TreeThinker 2026.04		18.54	22.51	20.07
TreeThinker 2026.04		17.29	23.65	19.74
SpreadsheetAgent 2026.04		16.04	22.22	18.42
Qwen3-235B 2026.04		15.33	22.51	18.09
GPT-4o 2026.04		10.52	17.66	13.27
TreeThinker 2026.04		10.16	19.09	13.6
Qwen3-30B 2026.04		9.45	18.23	12.83
GPT-3.5 2026.04		2.5	9.97	5.37
Mixtral-8x7B 2026.04		2.32	3.71	2.85
Llama-3-70B 2026.04		0.71	7.14	3.18
Binder 2026.04		0	0	0