CSV Parsing in 2026: Parsers Growing 4–6×, JSON Converters Flat

Every claim on this page is one HTTP request from being wrong. The window is 2024-10-01 to 2026-10-10, pulled from the npm downloads API on 2026-10-12 from this machine. "First" is the first 30 days, "last" is the final 30. Swap any package name at the end of that URL and re-check me — please do.

The table

Installs, not users. A CI cache miss is a download; one monorepo can emit dozens per developer.

The shape matters more than the percentages

I pulled monthlies to check whether any of this is a step function — a framework vendoring a parser and changing the count overnight. The honest answer is "not a step, but not a steady slope either." Here is every month of papaparse, no selection: 13.3M → 13.9M → 15.8M → 15.7M → 16.1M → 17.1M → 18.7M → 18.1M → 19.6M → 21.3M → 20.7M → 20.9M (2025), then 25.0M → 27.8M → 36.3M → 42.3M → 45.7M → 49.0M → 58.8M → 62.1M → 64.9M (2026). Quarters: +19%, +14%, +15%, +12% through 2025, then +42%, +54%, +36% in 2026.

I want to flag what I nearly published here. An earlier draft of this section read "21M → 21M → 25M → 28M → 36M → 42M → 49M → 59M → 65M: roughly +20% a quarter for two years." Those are all real months — I picked them. The ones I dropped are the ones where growth paused (Nov 20.7M, Dec 20.9M, May 45.7M, Aug 62.1M), and dropping them turned a lumpy 2025 plus a sharp 2026 into a smooth compounding line. A series is not falsified by containing true values; it is falsified by which ones you left out.

That distinction is the whole story, and the 2026 break is the part worth acting on. Through 2025 both readers grew ~11–15% a quarter — infrastructure drift. From January 2026 fast-csv went +46%, +69%, +61% and papaparse +42%, +54%, +36%, on a base neither had ever moved at before. Two packages that compete with each other inflecting in the same quarter is not one vendor's adoption curve; it is the input growing: more pipelines, more scraping, more exports, more agents whose tool calls emit CSV-shaped text. I cannot attribute the break to a cause with this data, and three years of steady drift suddenly quadrupling is exactly the kind of thing that deserves a mechanism I do not have. What I can say is that the acceleration is in both series, in the same quarters, and it is not a measurement artefact — the window is a full month in every one of those rows.

Where the growth actually concentrates

The spread is the second finding. The four packages that read CSV grew +286% to +784%. The ones that specialise grew far less:

• csv-stringify (write only): +187%
• csvtojson (a shape converter, not a parser): +59%
• csv (the kitchen-sink package that does both): +54%
• neat-csv: +9%, off a tiny base

My reading, held loosely: reading is the growth market and converting is not. A pipeline that ingests arbitrary sources wants a parser and then decides its own output shape; the fixed "CSV becomes JSON" middleman is exactly the job every parser now covers with three lines of user code. csvtojson's flat line is probably not declining usage of CSV-in-JSON-out — it is that job dissolving into the parsers. The data is consistent with that and cannot prove it; I am not going to pretend otherwise.

One confound I could not measure: d3-dsv rides inside the D3 ecosystem, whose install patterns follow whichever charting stack is fashionable. Its +561% may describe dataviz trends more than CSV trends. Exclude it and the reading-parser story still holds at +784/+490/+286.

What this site's parser is for

All of the above is Node packages that a build step installs. This page lives on a tool that is the opposite: no install, no dependency, no update stream, nothing leaving your browser. That is when a page like this one wins — csvformat.net is the right tool for forty rows you are pasting once. Papa Parse is the right tool for ten million rows you process nightly.

What this tool's parser handles (from the About page, all reproduced against the shipped code): quoted commas, "" escapes, newlines inside quoted fields, CRLF, ragged-row counting. What it consciously gives up: field trimming happens even inside quotes, only comma is a delimiter, and JSON output is all-strings so 01234 survives the trip.

The libraries in the table above mostly keep whitespace intact where this tool trims it — the opposite trade, and if padding is load-bearing in your data, that is another reason the library is the right tool there.

Decisions this does and doesn't settle

• Choose a streaming parser for anything you cannot hold in memory — the growth curve says you will eventually have to. The download counts here say nothing about performance, memory, or correctness; none of these packages has been benchmarked on this page, and I would not trust a page that claimed they had.
• Skip csvtojson-shaped dependencies in new code. Three lines over any parser above do it, and one fewer thing in the graph that needs updating in 2029.
• Do not read +784% as "CSV is winning the format war." Parquet, Arrow, and JSONL all eat CSV's lunch on paper. But exports, integrations, and humans-with-spreadsheets are the dominant CSV producers, and none of those flows is shrinking. Installs are a floor on usage, never a ceiling on fashion.

Figures are 30-day windows from the npm API as pulled 2026-10-12; the window slides, so re-pull anything load-bearing. October 2026 is a partial month in the monthly series (26.4M against September's 64.9M) and dips accordingly — that is measurement artefact, not collapse, and it is why the table uses 30-day windows instead. The quarterly growth rates above stop at 2026Q3 for the same reason.