Polymarket data download
Download Polymarket data: order books, prices and resolved markets
Files first. Four of the six below need no account at all, including a 897,192-snapshot research dataset published under CC BY 4.0 with a DOI.
All links checked on 2026-08-24.
Start here
What you can download
| Dataset | Size | Format | Access |
|---|---|---|---|
| BTC 5-minute order book sampleEvery 1-second snapshot of a single resolved market, with full ladders. The quickest way to see the shape of the data. | one full market lifecycle | CSV + JSON | No signup |
| Research dataset on ZenodoThree contract lengths — 5m, 15m and 4h — with the winning outcome attached. CC BY 4.0 and citable by DOI. | 897,192 snapshots · 805 markets | Parquet | No signup |
| Hugging Face datasetFive-minute markets only. Loads directly with datasets.load_dataset. | 158,745 snapshots · 352 markets | Parquet | No signup |
| Kaggle datasetConvenient if you already work in Kaggle notebooks. | the same 5-minute capture | Parquet per coin | Free account |
| Provider comparisonCoverage, resolution and pricing across the providers in this space, as a flat file. | small | CSV | No signup |
| Query it directlyThe API serves the whole archive rather than a fixed export. The free tier queries at the same 1-second resolution as the paid ones, over a shorter window. | unbounded | JSON over REST | Free account |
Which format to take
- CSV if you are opening it in a spreadsheet or eyeballing a single market. Ladders come through as JSON strings inside the cell.
- Parquet for anything analytical. The ladders are typed list columns rather than strings, so no parsing step, and zstd compression keeps 897,192 snapshots to 43 MB. Loads in pandas, Polars or DuckDB.
- The API when a fixed export is the wrong shape — a different date range, different markets, or a live feed rather than a snapshot in time.
import pandas as pd
df = pd.read_parquet("updown_5m.parquet")
df[df.seconds_to_close <= 5] # the book five seconds before settlement
(df.best_bid >= df.best_ask).mean() # measure the crossed rate yourselfWhat you can get from Polymarket directly
Polymarket's own API is free and needs no key. It serves price history down to 1-minute buckets, resolved markets and outcomes, and executed trades — for every market, including ones that closed years ago.
What it does not serve is order book depth. The /book endpoint returns the book as it is now; nothing archives it, and it cannot be reconstructed after the fact. The full picture of what exists and what does not goes through this in detail, including a measurement of why replaying event archives fails.
FAQ
Where can I download Polymarket data for free?
Several places, none requiring payment. A single-market BTC order book sample in CSV and JSON with no signup at all. A 897,192-snapshot research dataset on Zenodo under CC BY 4.0 with a DOI. The same capture on Hugging Face and Kaggle as Parquet. Polymarket's own API also serves prices and resolved markets free of charge.
What format is Polymarket order book data available in?
The free single-market sample is CSV and JSON. The larger research datasets are Parquet with zstd compression, which keeps 897,192 snapshots to 43 MB and loads directly in pandas, DuckDB or Polars. The API returns JSON.
Can I download historical order books directly from Polymarket?
No. Polymarket's /book endpoint returns the current state of a book and there is no historical equivalent. Order book depth is not archived at any granularity, so it has to have been recorded live while the market was trading.
Do I need an account to download these datasets?
Not for the BTC sample, the Zenodo dataset, the Hugging Face dataset or the provider comparison CSV. Kaggle requires a free Kaggle account, which is their requirement rather than ours. Querying the API needs a free PolyOrderbooks account for the key.
How large is the full Polymarket order book dataset?
The published research dataset is 897,192 snapshots across 805 resolved markets and 8 coins, 43 MB as Parquet. The live archive behind the API is considerably larger — over 779 million order book snapshots across 819,000 markets — and is queried rather than downloaded whole.
Can I use this data commercially?
The Zenodo, Hugging Face and Kaggle datasets are CC BY 4.0: commercial use is fine, modification is fine, attribution is required. API access is governed by the plan terms.