What Hyperliquid Perps Actually Cost: 90 Days of Funding Rates and Fees, Measured

Hyperliquid

Every comparison of perpetual exchanges settles the cost question with a screenshot of a fee table. Taker 0.045%, maker 0.015%, done. That table describes the smallest part of what a position costs, and for anyone who holds longer than an afternoon it is close to a rounding error.

The number that decides what a trade actually costs is funding, and almost nobody publishes it as a measurement. So we did. This page reads 90 days of hourly funding prints from the Hyperliquid API across 20 markets, native and HIP-3, and states what holding cost in each one.

What holding cost, market by market

The column that matters is the first number: what a long position paid, as a share of its own value, over the full 90 days. A positive figure means the long paid the short. A negative one means the long was paid to hold.

Three things in that table are worth sitting with.

The memory and AI names were the expensive end. DRAM, MU, INTC, NVDA and SNDK cluster at the top of the cost list. That is what a crowded long looks like in funding terms: enough traders want the exposure that they will pay every hour to keep it, and the rate never gets a chance to normalise.

The two crude markets paid longs. CL and BRENTOIL are the only markets in the study where the 90-day sum came out negative, meaning shorts financed longs over the window. They are also the two choppiest markets in the set by a wide margin, flipping sign more often than any other market measured. Their median hour was still positive; the negative total came from a small number of violent negative episodes rather than a steady drift, which is a different trade from a market that pays you calmly.

Gold and silver paid the long side almost every hour. GOLD charged the long in 97.7% of the hours measured and SILVER in 96.3%. Neither is the most expensive market in absolute terms, but they are the most relentless: there was almost no window in which a long could sit without paying.

How long the rate stays pointed the same way

A funding rate that averages out to something small is not the same as a funding rate that costs you nothing, because the sign persists. Over this window, LINK held the same funding sign for 627 consecutive hours, which is just over 26 days. ETH managed 536 hours and BTC 468.

That persistence is the practical risk. A trader who opens a long into the middle of one of those runs pays for weeks without a single hour of relief, and no amount of fee optimisation touches it. At the other end, BRENTOIL and CL changed sign hundreds of times over the same 90 days, so a position there was never on the wrong side of the carry for very long, but it also never had a stretch it could rely on.

If you are deciding between two markets that look similar on the chart, the flip count and the longest run in the table above tell you more about what the trade will feel like than the average rate does.

The fee arithmetic, in dollars

Here is the part the screenshots do cover, worked out in money rather than percentages.

HIP-3 builder markets cost double the native rate because a share of the fee routes to the builder that deployed the market. Every equity, index and commodity perpetual on Hyperliquid is a HIP-3 market on trade.xyz, so that doubled rate applies to all of them. Our Hyperliquid fees explained guide covers the VIP tiers and the HYPE staking discounts that reduce these numbers for high-volume accounts, and the fee calculator will do the sum for a specific trade size.

What the 4% referral discount is worth in real terms

It is worth having, and it is worth being honest about the size of it. The discount is 4% off the fee, not 4% off the cost of the trade, and it stops after the first $25 million of volume. The ceiling column above is the whole of it: at base-tier taker rates that is $450 across a trading lifetime on native perps, or $900 if every dollar of that volume went through HIP-3 markets.

Set that against the funding table. A $10,000 long held in the most expensive market in this study for the full 90 days paid more in funding than the referral discount can save across its entire $25 million lifetime cap. That does not make the discount pointless, since it costs nothing to have and it compounds for anyone trading in size. It does mean that a trader choosing a venue on the strength of a fee discount, while ignoring where funding sits, is optimising the small number.

When funding overtakes the fee

For each market we worked out how many hours a position has to stay open before funding at that market's median rate equals one taker round trip. On the major native perps it is roughly three days. On the slower HIP-3 equity and commodity markets it is closer to twelve, because those markets have both a lower median rate and a doubled fee to clear.

That break-even is the cleanest way to think about which cost you are actually managing:

• Held for minutes or hours: the fee is the cost. Use maker orders where you can, take the referral discount, and the funding column barely registers.
• Held for a few days: the two are comparable, and the market you picked starts to matter more than the order type you used.
• Held for a week or more: funding is the cost and the fee is a rounding error. At this horizon the only meaningful decisions are which market you are in and which side of it.

Most cost comparisons between perp venues are written as though every trader lives in the first bucket. The funding data says otherwise.

How to read funding before you open a position

Every market page on this site carries the live funding rate, refreshed on load, alongside open interest and 24-hour volume. Before sizing a position, three checks are worth thirty seconds:

1. Which way is the rate pointing, and how far. A positive rate means longs are paying. The funding rates tool shows the current rate across markets in one view.
2. How long has it been pointing that way. A rate that has been positive for a week is more likely to stay positive than one that flipped an hour ago.
3. What the annualised figure implies at your holding period. An hourly rate that looks like nothing annualises into a number that will decide whether the trade works.

Our funding rates guide covers the mechanism itself, including why the rate exists and how it keeps the perpetual anchored to spot. This page is the measurement rather than the explanation.

Methodology

Every figure on this page derives from the Hyperliquid API's fundingHistory endpoint at api.hyperliquid.xyz/info, read on 25 August 2026. The window runs the 90 days from 27 May 2026 to 25 August 2026 in UTC.

The endpoint returns at most 500 rows per call, so the collection script pages backwards through each market until the window is covered. Each of the 20 markets carries 2,160 hourly prints, one for every hour in the window, with no gaps.

The cost of holding a long over the window is the sum of the hourly funding rates across all 2,160 hours, expressed as a percentage of position value. It assumes the position is held for the full window at constant notional; a position sized up or down over that period would pay a different amount. The percentage of hours the long paid is the count of positive prints over the total. The longest one-sided run is the greatest number of consecutive hours the rate held the same sign. The break-even hours figure divides the taker round trip by that market's median hourly rate, which is why it can be positive on the two crude markets even though their 90-day totals are negative: their median hour charged the long, and the negative sum came from outliers.

The fee schedule is the published base tier: 0.045% taker and 0.015% maker on native perps, and 0.09% taker and 0.03% maker on HIP-3 builder markets. The HIP-3 pair is what a deployer fee scale of 1.0 produces rather than a fixed schedule, and a meta call against the trade.xyz DEX on 25 August 2026 returned a scale of 1.0 for all 116 of its markets. Accounts at higher VIP tiers, and accounts staking HYPE, pay less than the base tier shown here.

Funding rates move constantly, so the figures above describe the window they were measured over and nothing beyond it. This page is refreshed on a rolling basis by re-running the collection script; the window dates in the table update with it, and the as-of dates in the headline and the table are the authoritative ones.

You may republish these figures with attribution and a link to https://hyperliquidguide.com/ecosystem/what-hyperliquid-perps-cost.