Crypto funding rates and open interest: the pair, the units, and the plumbing

Crypto funding rates and open interest are the two public series that describe leveraged positioning, and they are almost always read badly — not because the concepts are hard, but because the numbers arrive in incompatible units, from venues with different conventions, through a collection layer that quietly decides what is knowable. This article is about that second half: what the pair actually measures, and what the plumbing behind it allows you to say.

The mechanics of funding — why a perpetual needs it, how the rate is built, who pays whom — are covered in what funding rates say about positioning and, at more length, in funding rates explained. What follows assumes that ground and takes up the harder question: given four venues that disagree about intervals, units and even which fields they publish, what conclusion is actually supported?

What crypto funding rates and open interest each measure

Two quantities, and the distinction is the whole game.

Funding is a price. It is what leveraged exposure costs to hold right now, set by the market, settled in cash between position holders. Positive funding means longs pay shorts. It is a statement about demand for leverage.

Open interest is a quantity. It is the total notional of contracts currently open — how much leveraged exposure exists. It rises when a new buyer and a new seller create a contract between them, and falls when both sides close. Critically, it does not move when a position simply changes hands: if you sell your long to someone else who wants it, volume prints and open interest does not budge.

Price and quantity. That is why neither is interpretable alone and why the pair is. A rising funding rate tells you leverage is getting more expensive; it does not tell you whether that is because more of it is being demanded or because less of it is available. Open interest closes that gap.

The reading framework for funding against open interest — four states, and what each one implies about whether a move is being financed by fresh leverage or by short covering — is set out in full in the funding article, and there is no point repeating it here. What that article does not cover, and what causes at least as many wrong conclusions, is the unit problem.

Open interest is the number nobody normalises

Ask three venues for open interest on the same contract and you can get three quantities that are not the same kind of thing.

Contracts. Some venues report the raw count of contracts outstanding. This is only meaningful if you know the contract multiplier, and the multiplier differs by contract.

Base units. Others report open interest in the underlying asset: so many BTC. This is stable against price, which is a genuine advantage.

Quote currency. Others report it in USD notional, which is base units multiplied by a price — usually the mark price, sometimes the last price, occasionally an index.

That third convention is responsible for an entire genre of bad analysis. Open interest denominated in USD moves when price moves, even if not a single position has changed. A 10% rally mechanically prints a 10% rise in dollar open interest on an unchanged book. Anyone reading "open interest surged with price, new leverage is entering" off a USD series has, in the limiting case, observed arithmetic.

The correction is to hold open interest in base units when you are asking about positioning, and in quote currency only when you are asking about notional at risk. Those are different questions and they deserve different series. When someone shows you an open-interest chart, the first thing to establish is which one it is; the second is whether it is one venue or a sum across several, because a sum across venues with different conventions is not a quantity at all.

The same discipline applies to funding, and there it is better known: a rate quoted per 8-hour interval and a rate quoted per hour are different units, and comparing them raw produces a conclusion that is exactly backwards. The funding cost calculator does the conversion and applies it to a position size, so the answer arrives in currency rather than in basis points — which also makes visible the thing traders most often forget, that funding is charged on notional and not on margin.

Reading open interest against price

Since the funding-against-open-interest table lives in the other article, here is the complementary one: open interest against price. It is a reading framework describing how the two series relate mechanically, not a measured result and not a set of outcomes.

The distinction between rows one and two is invisible in price alone and is the main reason to bother with open interest at all. It is also where the two series become a risk input rather than a directional one: a rally financed by new leverage leaves behind a stock of positions with maintenance margins clustered at known prices, and that stock is exactly what a move in the other direction converts into forced selling. The mechanism, including why the forced order arrives at the moment the book is thinnest, is in liquidations and cascades and in how to read a level-2 order book. If you want the arithmetic for a specific position rather than the general shape, the liquidation price calculator will give it to you.

None of this times anything. Crowded positioning raises the conditional severity of a move against the crowd. It does not raise the probability that the move happens today, and funding can sit expensive for weeks while price grinds in the direction that is paying for it.

The collection layer decides what you are allowed to say

Everything above assumes clean series. They are not clean, and the shape of the dirt matters more than its quantity. Here is what our own collection looked like on 2026-08-17, measured against production at 10:25 UTC. We publish it because a positioning number without its coverage is not checkable, and because the gaps are structural rather than random.

Four venues are sampled — binance, bybit, hyperliquid, okx — across six symbols: BTC, ETH, SOL, XRP, DOGE, BNB. That was more than 60,000 rows at the moment of measurement, split evenly across the four venues. Two fields tell the story:

These are not scattered holes. They are entire columns, across the whole history. A single completeness check over those two fields reports exactly 75.00% filled — and across all five collected fields, 85.00% — hiding precisely the structure that matters, because one of those two zeros is a bug and the other is not.

Hyperliquid's missing next-funding-time is correct behaviour. Funding there accrues continuously; there is no settlement timestamp to publish because there is no settlement instant. A comparison that assumes every venue snapshots funding at fixed times, and that positions can therefore be closed just before the timestamp to avoid the payment, is wrong on this venue specifically. Okx's missing mark price is a gap in what we collect — and so is its index price, which is likewise present on 0.00% of okx rows. Any percentile or spread computation that needs a mark price cannot include okx, and saying so is cheaper than quietly serving a three-venue number labelled as four.

Then there is the failure mode that no completeness check catches at all. On binance's BNB perpetual, the funding rate is exactly zero on 1,864 of 2,530 samples — 73.68% (measured 2026-08-17 at 10:25 UTC), in three continuous blocks. At those same 1,864 instants, hyperliquid and okx are never zero, and bybit is zero on 14 of them. Column fill on that field is 100.00%. A perfectly filled column carrying a series that barely moves would, on a naive cross-venue comparison, mark binance as the cheapest place to hold that position — not because it is, but because the series is stuck. This is the general lesson: completeness and informativeness are different properties, and only the first one is easy to measure.

Two more collection facts that constrain what can honestly be displayed — read at one instant, 2026-08-17 at 10:25:24 UTC, because these tables are alive and every count in them is larger by the time you read this. Sampling cadence is 60 seconds, measured as both median and mode across more than 2,500 intervals, over 2,400 of them exactly 60 seconds; the largest gap between two consecutive samples, over the day and eighteen hours of continuous collection on record at that instant, was 61 seconds. The 300-second staleness threshold therefore sits above every gap we have measured: the cadence is what is measured, the margin taken on it is a choice, and the two are not the same claim. The positioning series runs on 300-second ticks — more than 500 of them, every one exactly 300 seconds — and until 2026-08-15 at 16:00 UTC it also ran hourly, 3,600 seconds across 500 intervals, with no other value on either cadence. And the liquidation collector, switched on 2026-08-15 at 16:22:19 UTC and never restarted since, had recorded more than 1,100 messages from bybit and zero from binance: the binance socket connected at 16:22:21 UTC and has not delivered a single message on that stream in the time since. A liquidation chart built on that data would be a bybit chart wearing a market-wide label.

Percentiles need a window, and a window needs history

The single most useful thing you can do with a funding series is stop reading its level and start reading its rank. An annualised rate of some double-digit percentage is unremarkable in one regime and extreme in another; the only reading that survives a change of regime is distributional. Where does today's value sit in this asset's own recent history, on this venue?

Which raises an inconvenient question: how much history do you have? On 2026-08-17 at 10:25 UTC, our perpetual-metrics table held 1.757 days — one day, eighteen hours and ten minutes. Not 30, not 90. The positioning series reaches further back, roughly 22.6 days, but that depth is a splice rather than a series: an hourly cadence covering 2026-07-25 20:00 UTC to 2026-08-15 16:00 UTC, and a five-minute cadence that opens fifteen minutes before the hourly one stops and carries the rest. Neither table is enough to say the word percentile about a 30-day window.

So we do not say it yet. The first day a 30-day window actually exists is 2026-09-14, derived from the earliest timestamp in the table — 2026-08-15 16:14:52 UTC — and that is when a 30-day percentile becomes defensible. It is not a feature launch date; it is the date on which the sentence becomes true. The alternative — computing a percentile over less than two days and labelling it "30-day" — is the kind of thing that is technically a calculation and practically a fabrication.

The same discipline governs the staleness banner. The cadence is measured; the decision to call a source late after five missed cadences — a 300-second threshold on a 60-second cadence — is a choice, and the banner prints the threshold next to the age rather than rendering a bare verdict, so that a choice is not mistaken for a measured bound. Small distinction, and it is the difference between a dashboard that can be audited and one that has to be trusted.

Where crypto funding rates and open interest sit in the tiers

Crypto structure is a large part of what we cover, and the tier boundaries follow what it costs to produce a number rather than what it costs to withhold one.

Observer is free and includes the raw series: the four-venue funding view, the venue coverage table above with its zeros visible, and spot quotes and candles. If all you want is to see funding across binance, bybit, hyperliquid and okx side by side with an honest account of what is missing, that costs nothing.

Operator, at $39 a month or $390 a year, opens the derived layer — the positioning series with the window it covers and the age of each reading, the sentiment wall and the daily brief. Percentile context is not part of it, and no tier sells one: the section above is the reason, and a paid row for a window that does not exist yet would be the exact thing this article argues against. It does not open a crypto alert: nothing watches a funding rate, a liquidation print or a positioning reading and messages you when it crosses a level, on any plan.

Principal, at $79 a month or $790 a year, opens the execution-side surfaces: the level-2 book wall, the maker desk, raw flow series, chart reads, the REST API and outbound webhooks. The complete grid, row by row, is on the pricing page. Annual billing is twelve months charged as ten.

What none of the tiers include: execution, portfolio management, or advice. We publish measurements of market structure. What anyone does with them is their own decision.

Frequently asked questions about crypto funding rates and open interest

What is the difference between funding rate and open interest?

Funding is a price — what it currently costs to hold leveraged exposure, paid between longs and shorts. Open interest is a quantity — how much leveraged exposure exists. Funding tells you how expensive the crowding is; open interest tells you how large it is. Read alone, each supports a confident wrong conclusion.

Does rising open interest mean the price will go up?

No. Rising open interest means positions are being opened, and it is direction-agnostic on its own: it rises when new longs enter and when new shorts do. Paired with price, it distinguishes a move financed by fresh leverage from a move powered by covering — which is a statement about how much fuel is behind the move, not about where it goes next.

Why can't I compare funding rates between exchanges directly?

Because the interval differs. A rate quoted per 8-hour period and a rate quoted per hour are different units, and the hourly figure is necessarily the smaller number for the same annual cost. Some venues accrue funding continuously and publish no settlement timestamp at all — hyperliquid, in our collection, returns no next-funding-time on 100% of its rows, for exactly that reason. Convert everything to the same period before you look at it.

Is open interest reported in dollars or in coins?

Both, depending on the venue, and the choice changes the series. Dollar-denominated open interest moves with price even when no position has changed, so a rally alone can print an open-interest increase. For positioning questions, use base units. For notional-at-risk questions, use quote currency. Never sum series across venues that use different conventions.

How much history do you need before a funding percentile means anything?

At least as much as the window you are claiming. On 2026-08-17 at 10:25 UTC our perpetual-metrics history was 1.757 days, so we do not display a 30-day percentile: the first date on which that window genuinely exists is 2026-09-14. A percentile computed over less history than its label claims is worse than no percentile, because it looks authoritative.

Which venues and symbols do you cover?

Four venues — binance, bybit, hyperliquid, okx — and six symbols: BTC, ETH, SOL, XRP, DOGE, BNB. Coverage is published field by field, including the two fields that are entirely absent on one venue each, because a coverage figure averaged across fields would have hidden exactly the structure worth knowing.

Do you offer crypto funding alerts?

No, and no plan sells one. Nothing in the product watches a funding rate, a liquidation print or a positioning reading and sends a message when it crosses a level, so there is no funding threshold to set and none is offered on any tier. What alerting we do run is on Polymarket trades, delivered on Telegram, filtered by trade size and market category — crypto is one of those categories, so a large trade in a crypto market can reach you that way, but a funding rate cannot. To be precise about what the counter on the front page means: 335,737 alerts have been raised since 2026-02-02 — raised, meaning a threshold was crossed (measured 2026-08-17 at 10:25 UTC). Delivered alerts are a much smaller number, 29,276 since 2026-02-03, and the two are not interchangeable.

Crypto funding rates and open interest are among the most honest series this market produces: a price actually paid, and a count of contracts actually open. The dishonesty, when it appears, is downstream — in a sum across incompatible units, a percentile over a window that does not exist, or a completeness check that reports 75.00% and hides a column of zeros.

Everything above is published with its measurement date and its coverage, and the free Observer tier opens the four-venue funding view and the coverage table without a card. Look at the zeros before you decide whether the rest is worth paying for.