Cellprose writes an Excel workbook as plain Markdown: values in a pipe table, formulas as fill ranges, one file per workbook. Models see the formulas, not just the values, so they answer trace and typed-in questions that CSV and MarkItDown get wrong. It renders as tables on GitHub.
Pre-registered run of 2026-11-11: 58 held-out workbooks, 3 repeats, every token priced as uncached. Against agents that open the .xlsx themselves, cost is mostly no difference. Full results and limitations.
A 12-month forecast from the benchmark, converted in this page by the JS writer. Switch views, and hover a cell or a formula line. Coloured bands are fill ranges; the striped cells are typed-in numbers sitting where formulas belong.
bytes of XML in the forecast workbook once unzipped. Formulas, cached values and styling attributes on every cell.
characters of Markdown, written in this page by the JS writer. One pipe table per region, then a formulas block of fill ranges.
to answer, with the formulas in view. An agent that opens the .xlsx with a shell takes 3.5–4.6 turns to reach the same accuracy (97.0–99.2% on the real-world benchmark set).
Someone overwrote formula cells with typed numbers. This is what the values look like. Click the cells you suspect, then reveal what Cellprose shows.
| A | B | C | D | E | F | G | H | I | J | K | L | M | N |
|---|
A recorded benchmark call, replayed at a fixed token rate. Haiku 5.5 read the forecast once as CSV and once as Cellprose.
One recorded call, picked to show the kind of miss values-only formats make; it is not a typical call. Token counts are the whole call (6 questions); the reasoning text was not saved. It was recorded on 21 Nov 2026, on the v0.3 text (Haiku 5.5, one call per format, not repeated). In the 11 Nov 2026 run on 30 real-world workbooks, Haiku 5.5 scored 62.6% on typed-in questions with CSV and 99.4% with Cellprose; over all question types it scored 66.7% and 96.5% (all models).
Accuracy on 30 real-world workbooks, by how the model got the workbook. Cellprose in the prompt answers in one call; the shell agents take several turns and are somewhat more accurate on some models.
Pick an edit to the forecast. Git can only say a binary file differs; the .sheet.md shows which lines changed.
Binary files a/forecast.xlsx and b/forecast.xlsx differ
Both versions of the .sheet.md were written just now, in this page, by the JS writer from the edited workbook. The edited cell and the cached total next to it are patched in the XML; nothing else changes.
# convert, then compare sizes cellprose encode model.xlsx cellprose stats model.xlsx
Working with an agent? /llms.txt indexes the format for it, and /llms-full.txt is the spec, the skill and these steps in one file.
A few emails a year: releases and benchmark results. You confirm by email first, and every email has a link to leave. Privacy.