Sourcetable claims lead in AI spreadsheet benchmark

Sourcetable says its new benchmark put the AI spreadsheet platform ahead of Microsoft Copilot and Google Sheets on real-world data tasks. The company argues the results show a gap between chat-style add-ons and tools built to fetch, process, and analyze live external data.

Why it matters: - Sourcetable is positioning its platform as an AI-native alternative to spreadsheet tools that add AI on top of legacy software. - The benchmark focuses on tasks that matter for connected, API-driven work, including live data access, file parsing, and code execution. - The results could sharpen competition around which spreadsheet products can actually handle end-to-end data workflows, not just formula help and basic charting.

What happened: - Sourcetable published results from its self-administered Sourcetable Benchmark on July 8, 2026. - The benchmark used 24 tests designed to measure real-world spreadsheet and data work. - Sourcetable scored 24 out of 24. - Microsoft Copilot scored 19 out of 24. - Google Sheets scored 17 out of 24. - Sourcetable said the benchmark and its raw results are public at the company’s announcement.

The details: - The benchmark gave every test a real XLSX input, a real XLSX output, and one correct answer. - An open evaluation script checked cell types, formulas, formatted values, colors, fonts and formula results within a tolerance of 0.001. - Sourcetable said the benchmark tested connectors and files, two areas that require an AI system to reach outside its own environment. - Microsoft Copilot and Google Sheets both scored 0-for-2 on connector tests. - Sourcetable scored 2-for-2 on connector tests. - Sourcetable scored 4-for-4 on file tests that involved PDFs, DOCX, PNGs and other common formats. - Microsoft Copilot and Google Sheets each scored 2-for-4 on file tests. - Sourcetable said the benchmark was run in Light AI mode, which the company described as its most conservative setting. - Sourcetable said more powerful modes exist and were not used in the test. - The company said Copilot and Google Sheets did not appear to have full access to the internet or code execution. - Sourcetable also said Google Sheets does not have basic web search.

Between the lines: - The benchmark is both a product claim and a critique of how major office software vendors have layered AI onto existing systems. - Sourcetable is arguing that the biggest difference is not interface polish, but whether a tool can connect to live databases, third-party services and external files. - The company’s emphasis on methodology is meant to separate its results from demos or marketing surveys. - The public test suite is also an invitation for Microsoft and Google to reproduce the results or challenge the scoring.

What's next: - Sourcetable is pressing Microsoft and Google to run the benchmark themselves and publish their own numbers. - The company says it will keep the benchmark, code and supporting results available publicly. - The broader test now is whether competitors respond by improving connectors, file handling and code execution, or by disputing the benchmark’s design.

The bottom line: - Sourcetable is using a controlled benchmark to argue that its AI spreadsheet platform is more capable than two of the biggest names in office software when the work depends on real external data.