Prediction Market Data in 2026: APIs, Aggregators, and Datasets
Where prediction market data lives in 2026: free venue APIs, historical data and its traps, and when a data aggregator or provider beats building it yourself.
Prediction market data is quietly becoming one of the most interesting datasets in finance: live, crowd-priced probabilities for elections, rate decisions, sports, and world events, updated tick by tick. In 2026 you can get a surprising amount of it for free, straight from the venues, and the rest from a small but growing layer of aggregators and providers.
This guide maps the whole landscape: what kinds of data exist, what each major venue gives away, where historical data lives (and its traps), and when a prediction market data aggregator or provider beats wiring up venue APIs yourself.
Key Takeaways
- Live prediction market data is mostly free: Polymarket serves prices, order books, and streaming updates with no API key, and Kalshi serves markets, trades, and candlesticks over REST with no account. Its WebSocket needs a signed key, and its own docs disagree over whether the order book endpoint does, so test that one.
- Historical data is where it gets hard: every venue keeps different depth in different formats, with quirks like Polymarket's coarser history for resolved markets and Kalshi's rolling three-month live window.
- Aggregators and providers exist because of fragmentation: a dozen venues, a dozen schemas, and no common market identifiers. Normalizing and matching markets across venues is the actual product.
The Five Kinds of Prediction Market Data
Prices are the headline: a YES contract at 62 cents is a live 62 percent probability, which makes price feeds a probability dataset about the future, something almost no other source provides. Order books show the depth behind the price, which is what separates a real probability from a thin quote. Trades are the tick-by-tick record of what actually executed. Candles (OHLCV) compress price history into intervals for charting and backtesting. And resolution data, which markets settled and how, is the ground truth that research and model training depend on.
Free Prediction Market Data APIs, Straight from the Venues
The best-kept secret of this space is how much the venues give away. There is no single prediction market API covering all of it, so a data stack really means deciding which venue APIs you read directly and what you do about the rest.
Polymarket data
Everything readable is free and keyless: market metadata from the Gamma API, prices and order books from the CLOB API, wallet-level positions from the Data API, and a streaming WebSocket market channel for live updates. Price history comes from a dedicated endpoint at intervals from one hour to the market's lifetime. The trap to know: once a market resolves, that history degrades to 12-hour granularity, so capture fine-grained data from markets you care about while they are live. Full details in our Polymarket API guide.
One separation to keep straight at the time of writing: Polymarket US, the CFTC-regulated exchange built on the QCEX acquisition, is a different venue from the global platform rather than a regional skin on it, and it documents itself separately at docs.polymarket.us. Do not assume a market or an identifier from one side shows up on the other.
Kalshi historical data
Kalshi is the strongest venue for structured history: candlesticks at exactly three intervals, 1 minute, 1 hour, and 1 day, with batch endpoints and event-level aggregation, readable over REST without an account. The trap is the rolling window: the documented target for live endpoints is roughly the last three months, and older data moves behind a dedicated historical namespace with a cutoff you should query before assuming a range exists. Full details in our Kalshi API guide.
The order book is the one piece to test rather than trust. Kalshi's documentation disagrees with itself at the time of writing: the market data quick start walks you through fetching an order book with no credentials, while the order book endpoint reference lists the signed access headers as required. The WebSocket carries no such ambiguity and wants authentication at the handshake even on channels serving nothing but public data. Budget for depth needing a key even where prices do not.
The other venues
Coverage varies wildly. Opinion serves price history at multiple intervals but requires an API key for every read. Limitless offers price history but its short-lived hourly markets accumulate little of it. SX Bet has no candle endpoint at all: history means rebuilding from raw trade records. This unevenness is the first reason aggregation exists.
| Venue | Free live data | Historical data | The catch |
|---|---|---|---|
| Polymarket | Prices, books, WebSocket stream, no key | Price history, 1h to lifetime intervals | Resolved markets drop to 12-hour granularity |
| Kalshi | REST prices, trades, candles, no account | Candles at 1m, 1h, 1d; older data behind /historical/ | WebSocket needs signed keys; order book docs conflict |
| Opinion | Requires API key for every read | Price history, 1m to max intervals | Highest-friction reads in the space |
| Limitless | Prices and books with key | Price history, configurable intervals | Hourly markets accumulate little history |
| SX Bet | Fully public REST, no key | Trade records only | No candle endpoint; build your own from trades |
Prediction Market Historical Data: How Far Back You Can Actually Go
Depth is the question that quietly decides most projects, and the answer belongs to each venue rather than to the space as a whole. Kalshi reaches furthest in a documented, structured way, which is why most backtesting here starts with its candles and its historical namespace. Polymarket will hand you a market's whole lifetime, but with the resolution penalty above, so its long history turns coarse exactly where the settled outcome makes it interesting. Opinion and Limitless are simply young, and Limitless markets that live an hour never accumulate much to look back on. SX Bet leaves you rebuilding candles from trade records before you have a series at all.
Book-level tick data is the request that gets turned down most often. Trade endpoints are genuinely tick by tick, in the sense that every printed execution is there, but a matching record of order book state through time is not something these venues hand over as a download, and some cap how far back you can page. Teams who need it run their own recorder from the first week of a project. It is unglamorous work, and it is the difference between having prediction market historical data and wishing you had started earlier, because what you did not capture usually cannot be bought back.
Prediction Market Data Aggregators and Providers
The venues' data is free, but it does not add up to a dataset. Every platform uses different identifiers, price formats, and schemas, and none of them know that their "Fed cuts in September" market is the same event as a rival's. Turning a dozen incompatible feeds into one usable dataset means normalizing formats, matching markets across venues, and recording everything continuously. That work is the product that aggregators and providers sell.
The current landscape is small. PMXT is the open-source route: a CCXT-style SDK that normalizes market data across ten-plus venues, free to run yourself. Predictefy, our own platform, is the terminal-and-data route: markets matched across a dozen venues, prices and books normalized into one schema, and cross-venue tools like live arbitrage detection built on top; we are the vendor here, so weigh that accordingly. And the cautionary tale is Dome, a unified prediction market API acquired by Polymarket in February 2026 and shut down that April, a reminder to weigh any provider's independence and staying power before you build on it.
Prediction Market Datasets for Research and Backtesting
People searching for a prediction market dataset are usually picturing one file they can download, and that is the thing this space still does not have. There is no canonical archive here the way there is for equities, and nobody sells a complete, clean, cross-venue history yet. What exists instead is four sources with different shapes. The venue endpoints above, which you page and store yourself. Open-source tooling such as PMXT, which ships archive snapshots next to its live SDK. Commercial providers, ourselves included, who sell normalized cross-venue history. And your own recorder, the only one certain to hold exactly what your research question needs.
The practical playbook most quant teams land on: pull what the venues offer, with Kalshi candles and the historical namespace as the sanest starting point, run a recorder in parallel from day one, and treat resolution rules as part of the dataset rather than metadata around it. A probability series means nothing if you misread what the market actually settled on, and two venues can word the same question differently enough to settle it differently.
How to Choose Your Data Setup
Building on one venue? Use that venue's API directly; it is free and the deepest source for its own markets. Comparing or trading across venues? You need normalization and market matching, so start from an aggregator, open-source or commercial, rather than writing five integrations. Doing research? Prioritize whoever gives you the longest clean history in your category, and start your own capture today. For the full tour of every venue's API and what each exposes, see our guide to the 7 best prediction market APIs and SDKs in 2026. Nothing here is financial or investment advice, and a probability feed describes what a crowd believes rather than what is going to happen.
Frequently Asked Questions
Is prediction market data free?
Live data mostly is. Polymarket serves prices, order books, and streaming updates with no API key, and Kalshi's REST prices, trades, and candlesticks need no account, though its WebSocket requires signed keys and its documentation currently disagrees with itself over whether the order book endpoint does. Costs appear at the edges: venues that gate every read behind a key, deep or normalized historical datasets, and cross-venue aggregation are where paid products live.
Where can I get historical prediction market data?
Start with the venues. Kalshi offers candlesticks at 1 minute, 1 hour, and 1 day plus a dedicated historical namespace for markets that settled before its cutoff, and Polymarket serves price history per market, dropping to 12-hour granularity once a market resolves. Both have depth limits and format quirks, so for long, clean, or cross-venue history you will end up combining venue data with your own recording or a provider's dataset.
Is there a prediction market dataset I can download?
Not one canonical file, no. There is no equivalent of a standard equities archive for prediction markets yet. In practice a prediction market dataset gets assembled from venue endpoints you page and store, open-source archive snapshots from tooling like PMXT, a commercial provider's normalized cross-venue history, or a recorder you run yourself, which is the only source certain to match your research question.
What is a prediction market data aggregator?
A service that collects data from many prediction market venues and normalizes it into one schema, ideally also matching identical markets across platforms. Aggregators and data providers exist because every venue uses different formats and identifiers, which makes cross-venue comparison impossible without that translation layer.
Is there a free prediction market API for data?
Several, but no single prediction market API covers every venue. Polymarket and SX Bet serve market data with no key at all, Kalshi's REST market data is open though its order book endpoint is documented inconsistently, and Opinion requires a key on every read. Cross-venue coverage means either wiring up each venue yourself or reading from an aggregator that has already done it.
Conclusion
Prediction market data in 2026 is abundant at the surface and thin underneath: live prices are free almost everywhere, while long, clean, cross-venue history is still something you assemble rather than download. Start with the venue APIs, record early, respect the resolution rules, and reach for an aggregator the moment your questions span more than one platform, because that is the point where normalization stops being a chore and becomes the whole job.
From here: the Kalshi and Polymarket guides cover the two richest free sources in detail, and if the reason you want this data is the gaps between platforms, our arbitrage guide shows what those gaps are worth.
