SpreadsheetBench 2 — Results
| Rank | Model | Score |
|---|---|---|
| 1 | kimi-k3 | 34.8 |
SpreadsheetBench 2
1 phaseActive
Second-generation spreadsheet manipulation benchmark testing formula construction, data transformation, and multi-step spreadsheet workflows. Metric: % task success.
Overview
SpreadsheetBench 2
Quick answer: SpreadsheetBench 2 is a second-generation benchmark testing AI agents on realistic spreadsheet manipulation tasks — constructing formulas, transforming data, and completing multi-step workflows within spreadsheet applications. Kimi K3 scores 34.8% as of July 2026, reflecting the practical difficulty of real spreadsheet task automation.
At a Glance
What it tests: An agent's ability to correctly manipulate spreadsheets — writing formulas, transforming data layouts, and executing multi-step workflows to reach a specified end state.
Why it matters: Spreadsheets remain one of the most common enterprise data-manipulation tools. SpreadsheetBench 2 targets this specific, high-value automation use case where current model performance still has significant room to improve.
Known limitations: As a "v2" iteration, the exact differences from the original SpreadsheetBench and specific task composition are not yet independently published outside its citation by Moonshot AI.
What SpreadsheetBench 2 Measures
SpreadsheetBench 2 evaluates an agent's ability to complete realistic spreadsheet manipulation tasks, such as constructing formulas, reorganizing or transforming data, and executing multi-step operations to reach a specified target spreadsheet state. It represents an updated, likely harder revision of the original SpreadsheetBench task pool.
Benchmark Specifications
| Field | Value |
|---|---|
| Task category | Agent / spreadsheet manipulation |
| Metric | % task success |
| Saturation | Low |
| Created by | Not yet independently documented |
How SpreadsheetBench 2 Is Scored
Agents attempt spreadsheet manipulation tasks, with the resulting spreadsheet state verified against the expected target output, producing an aggregate % task success rate.
State-of-the-Art Results
| Rank | Model | Score | Source | Date |
|---|---|---|---|---|
| 1 | Kimi K3 | 34.8% | Kimi K3 technical report | 2026-07 |
Score sourced from Moonshot AI's Kimi K3 announcement, July 2026.
SpreadsheetBench 2 on Benchgen
No Benchgen results yet — be the first to run SpreadsheetBench 2.
SpreadsheetBench 2 vs Other Benchmarks
| Benchmark | What it tests | Saturation |
|---|---|---|
| SpreadsheetBench 2 | Spreadsheet manipulation & formula tasks | Low |
| OfficeQA Pro | Office document question-answering | Low |
| BIRD-SQL Dev | Structured data question-answering (SQL) | Medium |
| CorpFin v2 | Corporate finance task automation | Low |
Run SpreadsheetBench 2 on Your Model
Benchgen lets you run SpreadsheetBench 2 against your own agent, tracking spreadsheet task success rates over time as you refine formula-generation and data-transformation capabilities.