Evidence / 02

Claims with a measuring tape.

These are observed results on a Jason development machine in Chrome, not promises for every device.

Hardware, browser version, storage speed, column shape, and data type distribution all affect results.

Largest measured file20.97 GB888 million cells
Measured first import3.9 min20.97 GB CSV
Measured cold open111 msmedian after import
Local project size0.795 GiB24.6× smaller than source

Recorded runs

RunSourceCellsFirst importCold openNotable result
Large CSV / A20.97 GB888M3.9 min111 ms median0.795 GiB project
Billion-cell / B11.78 GB1B174.6 s93 ms median0 dropped frames

Measured against DuckDB-WASM 1.33.1

Every local browser tool in this category is DuckDB-WASM underneath, so formats are not the difference — they all read the same ones. The difference is what happens when the data has to stay open. Same machine, same pinned desktop Chrome, OPFS database, cross-origin isolation, spilling configured and confirmed accepted.

WorkloadDuckDB-WASM 1.33.1Jason
Run a query across the fileNever materialised — streaming scanCompleted14.05 GB in 76 sNot the workload Jason optimises for
Open the file and keep it workableMaterialised — what a spreadsheet actually isFailedOut of memory at 1.76 GB, after 87 s11.78 GB held open, 1B cells

The obvious objection was tested rather than assumed. The probe told DuckDB to spill to disk and recorded that every setting was accepted; it still ran out of memory, and it took 87 seconds to do it on a file it scans in 9. Desktop Chrome. Edge, Firefox, and Chrome on Android are not measured.

How to read these numbers

First import includes parsing the CSV and writing Jason's local project. Cold open measures reopening that already-built project. These are deliberately reported separately: one is a conversion cost, the other is the recurring working experience.

What is not claimed

Jason does not claim that every 20 GB file opens on every machine, that all datasets compress by the same ratio, or that all browsers reproduce these timings. The Free plan is also capped at exactly 1,000,000,000 source bytes.

Measure it with your own file.

Open an oversized CSV for free. Pro removes Jason's limits and adds structured-data, document, and multi-file Local Data Suite tools.

Open Jason free Inspect the benchmarks