Stock Market Data APIs

Most people arrive here needing one of three different things. If you want to pull prices into a script and pay nothing, start with Alpaca for real-time IEX or EODHD for worldwide end-of-day. If you are backtesting, the field narrows to providers that sell delisted securities — Norgate Data for prices, Sharadar for fundamentals — because a plain price endpoint silently hands you the survivors. If you are building a product other people use, the question is not price but redistribution rights, and it eliminates most of the listing before you compare anything else.

• Free, real-time, forever — Alpaca, IEX feed only.
• Worldwide end-of-day behind one key — EODHD from $19.99 a month, or Marketstack from $9.99 on a quota of 10,000 requests a month.
• Delisted US prices with index membership — Norgate Data.
• Point-in-time US fundamentals, delisted companies included — Sharadar, sold direct to individuals.
• Full tick and order book from exchange feeds — Databento.
• Tick history bought once and kept — Kibot or FirstRate Data for the US, Tick Data for about thirty equity markets worldwide.
• Financial statements rather than prices — Financial Modeling Prep or Intrinio.
• Raw options ticks beside equities — Massive or Databento; options-first vendors have their own section.
• Research on your own laptop, nothing shipped — yfinance, or OpenBB, which ships no data of its own but puts SEC, FRED, Cboe and fourteen more free sources behind one Python call and its own MCP server.

Three jobs, and why a price comparison across them says nothing

Everything here returns market data over an API, and that is close to the last thing these products have in common. A dollar figure means something different in each of the three.

Real-time is a live feed — a WebSocket you stay connected to, or a quote endpoint you poll for a price that is current rather than fifteen minutes old. This is the job where the word "real-time" does the most work and means the least, so pin down which venues are in it. Alpaca, Finnhub, Massive and Twelve Data sit here. So does IQFeed, which is not a web API at all: ticks arrive through a local Windows program over TCP sockets, and every real-time exchange bills its own fee on top of the $108.15 monthly core service. Barchart OnDemand is the breadth case — seventy-odd REST endpoints, with grain cash bids, USDA grain prices and refined fuel prices beside the equities — sold as a custom package with no published price.

Historical is bars and ticks in bulk, for backtesting and research. Judged on depth, on whether delisted securities exist at all, and on delivery: if you intend to backfill years of ticks, flat files are not a nice-to-have, because paginating that volume over REST is a project of its own.

The historical vendors separate on one fact each. AlgoSeek leases lossless capture of the SIP, OPRA and CME feeds — equities from 2007, derivatives from 2014 — on a per-asset-class rate card it actually publishes. Kibot sells US minute bars from 1998 once, as gzipped CSV you keep; FirstRate Data does the same in zipped CSV from 2000, with roughly 7,000 delisted tickers in the stock bundle. Tick Data sells by the symbol-year under a perpetual licence, with futures trades from 1974 and equity markets well beyond the US. CSI Data is end-of-day only and has no API, but its futures go back to 1922 and it builds a continuous contract eight different ways. At the cheap end, Tiingo sells adjusted daily bars back to 1962 from $30 a month.

Fundamentals is financial statements, ratios and filings rather than prices — a different pipeline, a different update cadence, and usually a different bill. Here the vendors split on who reads the data and how it arrives. Financial Datasets parses US filings straight from EDGAR and serves them to agents over REST and a hosted MCP server, with no free tier — $20 of credits is the floor. ROIC.ai sells statements and precomputed ratios for 70-plus exchanges over REST, MCP and Google Sheets, from $29 a month for five years of history and $89 for all forty-plus. SimFin ships its US statements as bulk zipped CSV datasets rather than call by call, with an as-reported variant for backtesting. Sharadar is the point-in-time case, below. The data itself is discussed at more length in fundamental data and research platforms; what lives here is the API in front of it.

Options used to be a fourth job here, and on 3 October 2026 it moved out to options data and flow analytics. It is its own market with its own economics: OPRA is the largest message feed in US equity markets, greeks and implied volatility surfaces are computed products rather than observations, and the history is sold by dataset, date and symbol instead of by monthly tier. The vendors bought for options first — ORATS, ThetaData, Cboe DataShop — are primary there; the general APIs that sell an options plan beside equities, Massive and Databento, stay here and are listed there too.

What used to be another job here was called alternative, and it was two unlike things wearing one word. The half that can be checked — Form 4, 13F and congressional filings — moved out to insider, 13F and congressional trade trackers, because it is bought by somebody who is not writing code and is judged on how fast a filing arrives rather than on a rate limit. The half that cannot is satellite imagery and card-transaction panels, which have no price page to compare and are not sold to individuals at all.

Ranking a tick archive against a free quote endpoint produces noise. Decide which of the three you are here for, then compare only inside it.

Six things to check before you read a pricing page

Vendors describe all six inconsistently, and a tier that looks cheap almost always differs on one of these rather than on price.

Latency. Real-time means an exchange feed. Delayed usually means fifteen minutes. End-of-day means a daily bar after the close. A provider calling a fifteen-minute quote "live" is not lying by its own definitions, which is exactly why the word is worth pinning down.

Source. Direct from an exchange, consolidated from the SIP, resold from another vendor, or aggregated from public pages. This is what determines whether full tick history exists at all, and it predicts price better than any tier name. It is also the difference between a product with a contract behind it and one that works until the upstream site changes — the reason yfinance is in this listing with a warning rather than at the top of it. The free scraped routes are also the ones that close: in 2026 Stooq put its CSV downloads behind a proof-of-work check and a captcha, and pandas-datareader 0.11.0 dropped its Stooq, Yahoo and Tiingo readers — replacing yfinance in a script is the way off them. Free routes with an API behind them close too: Nasdaq Data Link, which is what Quandl became, flagged one product of 210 as free in September 2026, and the WIKI US prices most people came for stopped updating in 2018 (what replaced Quandl).

History depth. Usually shorter than the marketing page implies, and usually shortest on the cheap tier. Intraday history is the field that disappoints most often: Massive gives 5 years of stocks history at $29 and 20-plus at $199 for the same API.

Free-tier limits. "Generous free tier" is not a unit. Find the requests per minute and per day, and whether the free tier is delayed or end-of-day — it almost always is. The caps are denominated differently by every vendor and are not comparable at a glance: 25 requests a day at Alpha Vantage, 20 at EODHD, 100 a month at Marketstack, 60 a minute at Finnhub, 50 an hour at Tiingo.

Redistribution. Whether you may show the data to your own users. Separately priced or forbidden outright, and the most common licensing surprise in the category. It is enforced through tier names, not a checkbox — Tiingo's $30 plan is personal and internal use only and the $50 one covers one organisation and two developers; Twelve Data's free tier is internal non-display; Marketstack's free tier is non-commercial.

Delivery. REST, WebSocket, flat files. A REST endpoint is fine for a dashboard and wrong for a backfill; a WebSocket is the only sane way to follow hundreds of symbols at once; flat files are what you want the day you load a year of ticks into a store of your own.

Survivorship bias is bought, not configured

No API fixes this for you and none claims to. A normal price endpoint returns what is listed today, so every bankruptcy, acquisition and delisting is absent from the universe you are testing against, and any screen or ranking run across history is run across the survivors. Correcting it requires a delisted-securities archive plus point-in-time index membership — someone has to have been recording the constituent list every day, and that recording cannot be reconstructed after the fact.

That is why it is a purchase rather than a setting, and why it is routinely the largest line in a serious backtesting budget. What you buy depends on what the backtest reads. For prices, Norgate's Platinum tier at $630 a year carries delisted US securities back to 1990 and historical constituent membership for the S&P, Russell, Nasdaq and Dow families — but its fundamentals are a current snapshot with no history. For financials, Sharadar is point-in-time: its as-reported figures are indexed to the date each filing reached the SEC and never rewritten by a restatement, and delisted companies sit in the same tables from the late 1990s. Bought direct, Fundamentals is $19 a month for five years or $39 for full history, Prices starts at $9, and the only constituent history is the S&P 500's. The direct store sells a personal-use licence and nothing else — a firm buys through Nasdaq Data Link, at no published price — and Sharadar puts its own delisted coverage at 99%, not 100%.

The other history vendors vary more than their pages suggest. FirstRate Data carries delisted tickers but says coverage thins with age; Kibot's own FAQ answers the question with "partially"; CSI Data keeps delisted stocks off its order form, as an add-on arranged by email at no published price. Backtesting on a survivorship-free universe is the worked version.

What the money actually looks like

Five price shapes exist here, and mistaking one for another is how a budget goes wrong.

A monthly tier, by rate limit. The commodity middle of the market, and the tiers are drawn on calls per minute far more than on data. Alpha Vantage is the clearest case: $49.99 buys 75 requests a minute with fifteen-minute delayed US data, and real-time does not appear until $99.99. Finnhub runs $49.99 to $199.99, billed quarterly or annually rather than month to month, with non-US exchanges only at the top — and even there London is delayed and the rest are end-of-day. Twelve Data bills the same idea in credits per minute, from $29 on Grow to $999 at the top of Ultra.

A monthly tier, by data. EODHD separates end-of-day at $19.99 from intraday at $29.99 and fundamentals at $59.99, so you buy the dataset rather than the throughput. Financial Modeling Prep does the same from $22 a month billed annually, where the tier decides how many years and which regions. Eulerpool sells global fundamentals, prices and macro behind one key by request volume instead: 100,000 a month free for non-commercial use, commercial use from $99, and a hosted MCP server on every plan.

Per gigabyte, or per dataset. How exchange-sourced data is actually sold. Databento charges by the uncompressed byte for history with no subscription at all, and its per-dataset subscriptions start at $199 a month and reach $4,500 for CME Globex with full depth. AlgoSeek leases by dataset or asset class at printed rates — $1,500 a month for one equity dataset on a twelve-month minimum, $5,000 for all four asset classes on a two-year term — so its real entry is $18,000 a year, and the raw files are deleted when the term ends. Cboe DataShop prices OPRA history per file, by date and symbol. This is the shape that has no free tier and no surprise: you know the number before you spend it.

Per year, rented. The backtesting corner. Norgate runs $270 to $787.50 a year depending on how far back the history goes and whether delisted securities are in it. CSI Data's thirty-year Gold packages are $639 to $1,889 a year plus a one-time licensing fee. Both are rentals: Norgate's database locks when the subscription lapses, and CSI revokes the licence after two months without an update.

Once, and kept. Intraday history you buy and keep rather than rent. FirstRate Data is $599.94 for the stock bundle, delisted names included, and $719.94 for 2.4 TB of trade and quote ticks from 2010. Kibot's bundles are $990 for every US stock and ETF in minute and daily bars, $2,400 with futures, forex and indexes added, and $4,800 with ticks. Tick Data prices by the symbol-year — $32 for US equity Level I, $180 for futures — against a $1,000 minimum for a new client. What you keep is the files, not the freshness: updates stop after a month at FirstRate (three for the tick archive) and after nine months at Kibot unless you subscribe, and Tick Data sells them as a separate annual subscription. None of the three licenses passing the raw data on.

Outside all five sit the vendors that print no number at all: Barchart OnDemand builds a custom package from a contact form, and Nasdaq Data Link's exchange feeds and premium datasets end in a Contact Us button.

The pattern across all five: the cheap end of this market sells you access, and the expensive end sells you the exchange's own bill passed through. Between them sits the only question that matters for a commercial product — whether the licence lets the data leave your machine.

Where this category connects

The API is one link in a chain, and the decisions either side of it are made elsewhere. A year of ticks has to land somewhere, which is tick data storage. The read pattern you optimise for is usually decided by your backtesting framework. If the data is going into an LLM rather than a script, most of these vendors now publish their own server — see financial MCP servers. And if what you actually wanted was a chart and a screener rather than a key, that is charting platforms and screeners.