Surge AI launches GDP.xlsx: 70 real-world tasks testing agents on messy spreadsheets
echen · x · 2026-10-01
echen announces GDP.xlsx, the sibling to GDP.pdf: a Surge AI benchmark of 70 real-world tasks across 12 domains (finance, healthcare, insurance, legal, logistics, etc.) testing whether frontier models can navigate professional Excel workbooks.
- Real spreadsheets are small software systems: a needed number may depend on a lookup table on another tab, colors encode business rules, a comment in Sheet2!D18 can change how a calculation is interpreted.
- Models must inspect workbooks with tools, cross sheets, examine formulas and formatting, and answer questions a colleague would actually ask.
- The hard part is reconstructing workbook logic: which source is authoritative, which summaries are stale, which exceptions and approval states apply.
Related event: Surge AI Releases GDP.xlsx Benchmark; Top Agent Scores Just 38.3%(2 posts)→
More from Research
- Mathematician details agentic math workflow with Codex CLI and Claude Code, results coming — burny_tech · 2026-10-01
- Egocentric video dataset EgoPro from LightwheelAI trends on Hugging Face — LightwheelAI · 2026-10-01
- NVIDIA Opens 2027–28 Graduate Fellowships With $60,000 Awards, Due Oct 30 — NVIDIAAI · 2026-10-01
- Airbench Crowdsources a Local LLM Leaderboard via One-Prompt Agent Benchmarks — dh7net · 2026-10-01
- Five UW Allen School Faculty Win NSF CAREER Awards Spanning Quantum Crypto to Brain-Inspired AI — lazowska · 2026-10-01
- GDP.xlsx benchmark: 70 real spreadsheet tasks, best frontier agent scores only 38.3% — echen · 2026-10-01