Polymarket Historical Data: Where to Download It (2026)

What historical data does Polymarket publish itself?

Polymarket runs public APIs that anyone can read. The Gamma API serves the market catalog and event metadata, and the CLOB API serves order books, trades and price history for individual markets. Both are documented and neither requires payment to read.

What they are built for is current state. You can ask what a book looks like now, and you can pull price history for a market. What you cannot do is ask what the full book looked like at 14:32:07 on a Tuesday in March, because nobody serves an archive of every book mutation after the fact. That data existed at the time. Preserving it is a recording job, and it has to have been done while it was happening.

That is the whole reason a market for Polymarket historical data exists. The providers below are, with few exceptions, companies that started recording at some point and are selling access to what they captured since.

One practical note if you are reaching for Polymarket's own Python client: py-clob-client is the most-starred option and it is archived by its maintainers, with its last commit in May 2026. It still installs and reads fine, but there will be no patches when the API shape moves. Check the current state of the clients before you build on one.

Where can you download Polymarket historical data?

The split that matters is in the second column. If you need prices over time, most of these will do and the cheapest option wins. If you need the book with size at each level, the field narrows sharply and the capture start date becomes the binding constraint, because no provider can sell you what they were not recording yet.

How do you pull Polymarket candles with an API?

Candles are outcome keyed rather than market keyed, which is the first thing that trips people up. A binary market has one outcome per side, so you resolve the outcome before you ask for its history.

curl -s "https://data.predictefy.com/api/polymarket/fetchOHLCV?outcomeId=OUTCOME_ID&resolution=1h&limit=500" \
  -H "Authorization: Bearer pk_live_YOUR_KEY"

limit tops out at 5,000 candles per call, so a long range is a paging job using start and end rather than one enormous request. Both accept an ISO timestamp or epoch milliseconds.

An empty response is not an error. It means the requested range predates available capture, contains no trades, or crosses a capture gap. When history genuinely cannot be served the route answers a retryable 503 HISTORY_UNAVAILABLE, which is a different thing and should be retried with backoff.

What is the difference between a true candle and a derived one?

This is the question almost nobody asks and it decides whether a backtest means anything.

A true venue candle is one the exchange itself computed from its own trades. A derived candle is one somebody reconstructed, by rolling up finer candles, by deriving from a trade tape, or by writing forward from recorded catalog prices. All four look identical in a chart. They are not equally trustworthy inputs to a strategy.

Every candle Predictefy returns carries four fields that say which it is:

{
  "timestamp": 1780444800000,
  "open": 0.61,
  "high": 0.63,
  "low": 0.6,
  "close": 0.62,
  "volume": null,
  "source": "write-forward",
  "sourceType": "point-derived",
  "quality": "ok",
  "isTrueCandle": false
}

source is where it came from: official from a venue history API, onchain, write-forward from a recorder, or derived. sourceType says how it was built, with rollup meaning query-time aggregation of stored finer candles. quality reports ok, partial, suspect or mixed, where partial flags a known capture gap inside the bucket. And isTrueCandle answers the question directly.

Read those before you trust a number. A quality of partial on the candles underneath a backtest is the difference between a result and a story.

Can you get historical order book depth, not just prices?

Yes, and it is a different and more demanding product than candles.

A candle tells you where the price went. A historical book tells you what size was available at each level while it went there, which is what decides whether a strategy you are backtesting could have actually filled. A strategy that looks profitable against midpoints and unprofitable against real depth is not a strategy, and you cannot tell the two apart without the book.

The providers differ on one technical decision that matters more than their pricing: sampling method.

Snapshot on change records a new state every time the book mutates. Nothing is smoothed and nothing is interpolated. Probalytics works this way, recording Polymarket from November 2025.

Fixed interval records the book every second or every minute regardless of activity. Simpler and smaller, and fine for slower questions, but it averages away exactly the fast movement that microstructure work is about. PolyOrderbooks captures at one-second resolution and PolyBackTest at one minute.

If your question is "could I have filled 500 contracts at that price", you want snapshot on change. If it is "roughly how deep was this market last quarter", interval sampling is enough and cheaper.

How far back does Polymarket historical data go?

It depends entirely on the provider, because coverage begins when that provider started recording rather than when Polymarket launched. Probalytics states Polymarket coverage from November 2025. Others differ, and several do not publish a start date at all, which is worth asking about directly before you buy.

For candles, the picture is different by resolution. Minute, hour and day candles are stored and reach back furthest. Sub-minute candles are forward only from the start of capture coverage, because a one-second candle cannot be reconstructed from data nobody recorded at one-second granularity.

The practical approach is to probe rather than assume. Request the exact range you need with start and end, look at what comes back, and check the quality field on the edges of the range where gaps are most likely.

How much does Polymarket historical data cost?

Reading Polymarket's own public APIs is free, within their rate limits. Archives are where the cost is, and it scales with how much fidelity you want.

On Predictefy, history is metered in credits rather than sold as a dataset, so you pay for the range you pull:

The free plan carries 25,000 credits a month, which covers a substantial amount of exploratory pulling before anything is owed. The specialist archives mostly run free tiers with published pricing pages above them, with the exception of TickFoundry, which is paid and waitlisted.

Frequently Asked Questions