Skip to content
Open format · .sheet.md

Spreadsheets that models read right, in one call.

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.

93.8–99.3%right on formula-heavy questions in one call, against 65.9–84.1% for CSV and 70.0–85.7% for MarkItDown (30 real-world workbooks, four models)
1 turnwhere an agent with a shell needs 3.5–4.6 turns (97.0–99.2% accuracy; Cellprose is within about 0.5 to 3.8 points of it per model, GPT-6 Luna 93.8% against 97.6%): no sandbox needed
0.32–0.60×MarkItDown's cost per correct answer, a win in 7 of 8 model-and-corpus pairs

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.

See it

Every formula, in ten lines.

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.

How it works

From a zip of XML to text a model can read.

1 · THE .XLSX—

bytes of XML in the forecast workbook once unzipped. Formulas, cached values and styling attributes on every cell.

2 · CELLPROSE—

characters of Markdown, written in this page by the JS writer. One pipe table per region, then a formulas block of fill ranges.

3 · THE MODEL1 turn

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).

Try this

Find the typed-in numbers.

Someone overwrote formula cells with typed numbers. This is what the values look like. Click the cells you suspect, then reveal what Cellprose shows.

ABCDEFGHIJKLMN
0 cells picked
Replay

Same question, two formats.

A recorded benchmark call, replayed at a fixed token rate. Haiku 5.5 read the forecast once as CSV and once as Cellprose.

Workbook 11 · Haiku 5.5 · repeat 2 · one question of 6 in the callConsider the headcount and payroll rows for Feb through Dec (C4:M5). List the cells that hold typed-in numbers rather than formulas.
Cellprose used 237 more output tokens (10% more) on this call and found both gaps; CSV missed one.

CSV

Read811
Reasoning + answers2,406
["Forecast!K5"] Missed H4, which holds the value its formula would give.

Cellprose v0.3

Read1,054
Reasoning + answers2,643
["Forecast!H4","Forecast!K5"] Correct: both gaps in the fill ranges found.

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).

Benchmark

Close to agent accuracy, in one call.

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.

CSV1.0 turn · $0.0972 per correct
82.7%correct
MarkItDown1.0 turn · $0.1286 per correct
81.4%correct
Cellprose in the prompt1.0 turn · $0.0567 per correct
99.3%correct
Agent + shell on .xlsx4.6 turns · $0.0841 per correct
98.7%correct
Agent + formula viewer4.7 turns · $0.0792 per correct
97.3%correct
30 real-world workbooks, 163 questions, 3 repeats, run of 2026-11-11 · cost at list price, every token uncached · verdicts, intervals and limitations
Works with Git

Review a spreadsheet change like code.

Pick an edit to the forecast. Git can only say a binary file differs; the .sheet.md shows which lines changed.

K5 held a typed 480,000. Put the formula back and the total follows.
git diff forecast.xlsx
diff --git a/forecast.xlsx b/forecast.xlsxBinary files a/forecast.xlsx and b/forecast.xlsx differ
git diff forecast.sheet.md0 lines changed
Running the writer…

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.

Get started

Use it where you work.

# 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.

Launch news

Hear when it ships.

A few emails a year: releases and benchmark results. You confirm by email first, and every email has a link to leave. Privacy.

Cellprose is an open format for spreadsheet workbooks. Benchmarks: pre-registered run of 11 Nov 2026, 58 held-out workbooks, four models, three repeats. Privacy

CellproseGitHub ↗
A1fx
Ready