SpreadSheetBench-v1Agents

SpreadSheetBench (NeurIPS 2024 Spotlight) is built from 912 real Excel forum questions and evaluates LLMs on practical spreadsheet operations. It includes multiple sheets, non-standard relational tables, and rich non-text elements, using an online-judge-style multi-test-case evaluation mechanism.

Ranking Sources

#ModelScore

1

MiniMax M3MiniMax

89.4%

2

Claude Opus 4.6Anthropic

89.3%

3

Claude Opus 4.7Anthropic

88.5%

4

GPT-5.5OpenAI

88.1%

5

Qwen3.7-MaxAlibaba

87.0%

6

Qwen3.7-PlusAlibaba

86.3%

7

GLM-5.1Zhipu AI

85.2%

8

MiniMax M2.7MiniMax

84.9%

9

DeepSeek-V4-ProDeepSeek

84.9%

10

Kimi-K2.6Moonshot AI

84.5%

12 models total

Back to benchmarks