Crypto data API quota calculator

What it computes

One job, priced six ways. The job is what you would write down before opening any pricing page: how many symbols, which candle interval, how often you fetch the new candles, how many days of history you load once at the start, and whether you also poll a latest price several times a minute. The calculator turns that into requests, then into each provider's own unit, and then sets it against two different limits, because every provider here has one or both of them.

The budget. A quantity that runs out: Binance request weight per minute, CoinAPI credits per day, CoinGecko, CoinMarketCap and Glassnode credits per month. The running job is converted into the budget's own period, so a per-day load is compared with a per-day quota and multiplied out to a month for a monthly one. Where it does not fit, the verdict says by how many times, rounded up.

The rate. A ceiling on how fast you may ask, whatever the budget says: calls per minute on CoinGecko, CoinMarketCap and Glassnode, and per-endpoint counters in requests per two seconds on OKX, which has no budget at all. The busiest minute is taken to be the one in which a refresh fires every symbol at once, plus that minute's ticker polls. A job that exceeds the ceiling can usually be paced out; the verdict flags it so that you plan the pacing rather than meet a 429.

The backfill is shown separately, because it happens once: its cost in the provider's unit, the share of one period's quota it uses, and how long it takes when paced at the rate limit.

Where the numbers come from

Two places, both already on this site and both dated. What one request costs in each unit comes from why API quotas don't compare, which reads it from each provider's own documentation and cites the page: Binance's weight per endpoint, CoinGecko's credit per successful call, CoinMarketCap's credit per 250 items on latest quotes, CoinAPI's credit per 100 points returned when a call uses limit, and Glassnode's credit per asset per request. The quota, the per-minute limit, the page size and the history each tier reaches come from the provider's card, each verified on the date the rate table prints beside it.

The default scenario is the guide's own worked job, a year of daily candles for 20 assets, refreshed once a day. Its backfill column should match the guide's table line for line; where it does not, one of the two pages is wrong, and the error report at the foot of this page is the way to say so.

The figures live in one table that both this form and the page's Markdown twin read, so the twin prints every number the form multiplies, with its document and date.

What it leaves out

Providers whose meter is not a formula. Dune charges for compute it prices only after a query runs, and its two billing pages disagree on what reading a result costs. CryptoQuant deducts by rows returned and data type and publishes no rate card. Nansen serves OHLCV from a backtesting endpoint group whose credit cost is not stated. Amberdata and Kaiko sell access per venue, market or instrument; the meter exists but is not what you pay for. The rest of the market data APIs are not in the guide's conversion and are not modelled yet.

Quota changes. Free tiers in this market are re-cut, often without notice: CoinMarketCap changed what a credit buys on 7 August 2026, and its own pages still disagree about the free allowance. The table carries the date each figure was read. Check the provider's page before relying on a verdict near the line.

Enterprise tiers and overage. Each provider is modelled at the cheapest tier that has an API: free for Binance, OKX, CoinGecko and CoinMarketCap, Startup for CoinAPI, which has no free plan, and Professional with its included credits for Glassnode. Higher tiers, purchased credit packs and CoinAPI's metered overage are not priced; "needs 4× the quota" is the question to take to the pricing page, not its answer.

Websocket streams. A stream is metered differently, by connection, by subscription or by bytes, and a job that streams candles instead of polling them does not spend REST weight or credits in the way modelled here. CoinAPI bills websocket data by the gigabyte and CoinGecko charges a fraction of a credit per websocket response; neither is in the table.

The page size. Where a card states how many candles one request returns, that figure is used: 1,000 on Binance and 300 on OKX, applied to OKX's history endpoint too because the card states no separate figure for it. Where a card states none, as for CoinAPI, the calculator assumes 1,000 and says so. A smaller real page means more requests, which matters for the rate limit; on CoinAPI's per-100-points meter it barely changes the credits.

Batching, per-IP sharing and failed calls. Each symbol is one request, except on CoinMarketCap where the card documents the batch rule for latest quotes. An endpoint that prices many symbols in one call would cost less than shown. Binance and OKX count per IP, so every process behind one address shares the budget. Retries are not counted, and on CoinGecko a failed call spends rate but not credits.

CoinGecko's candles are price points. On the Demo plan the historical chart endpoint returns prices at a granularity chosen from the range, not OHLC candles; it is what the guide prices and what is modelled here. Its OHLC endpoint returns 4-day candles beyond 30 days on Demo.