Free vs Paid Polymarket Data: What Each Source Actually Contains

Start with the question, not the price

The free-versus-paid decision in prediction-market data is not really a budget decision. Almost every dataset in this space answers one of two very different questions, and the cheap sources answer only the first one.

The first question is what happened: what did this market price at, when did it resolve, which way, and who traded it. That is last-price and metadata work, and it is well covered for free.

The second question is what would have happened to me: if I had sent this order at this moment, would it have filled, at what average price, and how much would size have cost me. That is order-book work. It needs the resting depth as it stood at the moment of the decision, and no free Polymarket source carries it.

What each free source actually contains

Why no free source has the historical book

This is a structural fact about the market, not a gap someone forgot to fill. An order book is a live object: it changes on every new order, cancel and trade, and once a moment has passed the state of the book at that moment is gone unless somebody was connected and writing it down at the time.

Polymarket streams the live book over its CLOB websocket but does not serve an archive of it. That means historical depth cannot be backfilled after the fact, at any price, by anyone who was not already capturing. A dataset published today can contain last prices going back years and still contain zero historical order books.

The practical consequence is that the usual escalation path — start free, upgrade when you outgrow it — does not work for order-book questions. The free tier is not a smaller version of the paid one; it is a different measurement.

The test that settles it in one minute

• Write down the exact question you need answered, in one sentence.
• Ask whether answering it requires knowing what size was resting at what price at a specific past moment.
• If no — prices, outcomes, volumes, wallet flow, market metadata — the free sources are genuinely enough, and you should use them.
• If yes — fill probability, slippage, spread capture, market-making, execution cost, liquidity over time — no free Polymarket source can answer it, regardless of file size.
• If you are unsure, check whether your backtest currently assumes it filled at the last traded price. If it does, it is answering the second question with the first question's data.

What a recorded archive adds, concretely

DepthFeed's Polymarket archive holds 450 million+ order-book snapshots across 380,000+ markets, captured event-driven from the CLOB websocket rather than sampled on a clock, reaching back to January 2026 on the longest-running series. Each snapshot carries both the exchange timestamp and DepthFeed's receive timestamp, so latency is measurable rather than assumed.

On the flagship BTC up/down 5-minute markets, the book typically shows a one-cent (single-tick) spread with around 49 price levels on each side and roughly 94,000 shares of resting depth. Those are the numbers a fill model needs and the ones a last-price series cannot supply, measured as of June 21, 2026.

An honest read on when not to pay

There are common cases where paying for prediction-market data is the wrong call, and it is worth naming them.

If you are studying forecast accuracy — how well market prices predicted outcomes — last-price history and resolution data are the correct inputs and they are free. If you are building a dashboard, tracking volume, researching wallets, or writing about the space, on-chain indexers and the official API cover it. If you are testing a slow strategy on multi-day markets where you would cross a one-cent spread once and hold to resolution, the execution detail may not change your conclusion enough to matter.

The cases where it does matter are short-dated markets, size that exceeds the top of book, anything that quotes rather than crosses, and any result whose edge is smaller than the spread you assumed away.