Axi

Axi measured for a month: 5th of 10 on annual cost

One instrument, one window, one profile of 10 lots a month, on an account we opened ourselves. Where Axi beats the broker we are paid by, that is printed too.

EUR/USD · 27.08–25.09 · our own account, 1 s bid/ask sampling

The bill, taken apart: $1 435 a year

The headline figure is one number, and one number is not checkable. Here it is in its parts, for 10 lots a month of EUR/USD on the Pro · MT5 account we opened:

One pip on one standard lot of EUR/USD is about $10; that conversion is the only constant in the sum, and it is the market's, not ours.

On this account the commission is the larger half of the bill: $840 of $1 435, or 59%. That is the thing a spread table cannot tell you. Axi quotes a raw spread of 0.45 pips, which looks almost free, and then charges per lot for the privilege. Whether that is cheaper than a wider spread with no commission depends entirely on how much you trade, which is why the profile is stated rather than assumed.

The swap line, $55 a year, is the one component that does not scale with how often you trade — it scales with how long positions stay open. A Flat overnight, you pay almost none of it; holding for weeks, you pay more than this profile implies.

The table is an invoice reconstructed from measurements, not a price list. Change the volume and every per-lot line moves with it. The per-lot components scale with volume and the swap does not, so the ordering between accounts is a function of how much you trade.

Spread by hour on this account

Axi has no publishable hourly breakdown. The hourly table needs a continuous series from our own account for the whole window, and one that survived the feed audit. This account either post-dates the window or failed that audit. An hourly table assembled from a partial series looks exactly like a complete one, which is precisely why it is not here.

What we do publish for this broker is the aggregate above, with the window it covers. When a full series exists, this section fills itself from the same store as every other figure on the site — there is no second pipeline for presentation, which is the only reason a reader can trust that an hourly figure and an average come from the same ticks.

The hours that matter most are the ones nobody advertises. Between 21:00 and 24:00 UTC the New York session closes, liquidity thins and spreads on every account we hold widen — several of them by a multiple rather than a margin. A broker quoting a single daytime average is quoting its best hour and leaving the arithmetic to you; the spread page carries the day-against-rollover ratio for every account in the table, including this one.

The rig behind Axi's numbers

An account was opened at Axi in the ordinary way, on the Pro · MT5 account type — the one a retail client is offered, not a negotiated institutional tier. The broker's own platform runs on a machine that does not sleep, does not hibernate and does not throttle on battery; all three defaults had to be turned off explicitly, because each would have cut exactly the overnight hours this site exists to measure.

Bid and ask arrive from that platform and are stored raw with the timestamp the server put on them. The server clock is not UTC — one broker in this set runs three hours from it — so the offset is measured from fresh ticks at every start rather than written down as a constant. A tick that is not seconds old after the offset is subtracted is not stored at all, which is the guard that closed the hole through which a frozen Friday quote once arrived stamped as Saturday.

Execution is measured the only way it can be — by putting orders through the account: 4 795 market orders of the smallest volume the account permits went through this account during the window of which 209 were timed into the minutes around scheduled economic releases. Each one was closed immediately - an open position would accrue swap and market exposure and poison the next measurement - and after every run the account was reconciled against the broker to confirm nothing was left open.

2026-08-27 to 2026-09-25, continuously, with coverage counted against calendar market hours rather than against the hours we happen to have seen.

Fills, rejects and slippage around releases

The spread is advertised the fill is not and the difference between them is where the money actually goes. Out of 4 795 orders on this account 0.7% did not execute and the 209 orders placed into the minutes around scheduled economic releases slipped by 1.2 pips on average — about $12 per standard lot every time one of those orders is placed.

Rejects are the broker's refusals, taken from its own status codes. Our own timeouts have a separate status and are not folded in — doing so would inflate a broker's rate with our latency. Keeping them apart is what makes the figure comparable between brokers at all.

Slippage measures where the fill landed relative to the request, with the sign pointing the way a trader experiences it — positive is money lost. It is measured only on release-timed orders because that is where the number has meaning. A grid of probes spread evenly through a quiet afternoon reports that every broker on earth is identical. The whole of execution quality lives in the few minutes a day when liquidity thins and everybody has orders arriving at once.

Each probe is the smallest volume the account allows and is closed immediately, which keeps the measurement from accumulating market exposure and keeps the cost of running it survivable. Where a fill and the surrounding quotes imply different scales, the probe is marked suspect — one of our feeds sends scaled integers, and mixing the two produced a plausible nonsense once already.

What Axi's licence gives a client

Not established. We have not confirmed which of this broker's licensed entities would hold an account like ours, and without that there is no honest way to say what protection applies: a group can hold a European licence, an offshore one and several in between, and the client protections attached to them are not comparable. A licence number attached to the wrong entity is worse than no licence number at all, so this section stays empty until the broker tells us which entity our account sits under.

This matters more than it sounds. The compensation ceiling, the maximum leverage you may be offered, whether your losses can exceed your deposit and who hears a complaint are all properties of the entity, not of the brand on the website. Two clients of the same broker, onboarded through different entities, can have materially different protection with identical platforms and identical spreads.

Where we have established the entity, this section carries the regulator's own rule with the sentence it came from — see any of the brokers whose entity is confirmed. For this one, the measurements above stand on their own: they describe the account we opened, whoever turns out to hold it.

Eight axes, eight placings for Axi

Axi is in the better half of the table on 7 of the 8 axes it has figures for — best of 9 on withdrawal fee. Each placing is against the other accounts in the same source class, on the same instrument and window - not against a broker's claim, and not rolled into a composite score. A composite score is exactly where a site paid by one of these brokers would hide its thumb, so there is none.

Lower is better on every axis · 27.08–25.09 · ranking happens inside one source class only

Read the column of placings rather than any one row. Consistency across axes tells you what kind of account this is; a strong placing on one axis tells you almost nothing about the bill or the fills.

Questions a reader usually asks next

Each answer is derived from the figures on this page. None of them is a general statement about brokers.

Auditing this page

A measurement published by a site that earns a commission from one of the brokers it measures deserves scepticism, so here is where to point it.

• Check that the losses are printedif the sponsor's losses ever stop appearing on this site, stop believing the rest of it
• Check the source class on each number🔬 measured, 📊 third party, 📄 stated by the broker — and no ranking crosses those classes
• Recompute the billthe components and the profile are printed above; the arithmetic is deliberately simple enough to redo on paper
• Ask for the rowsthe data page says what we hand over, including the flagged readings we excluded
• Compare us with anybody else measuring the same thinga figure that only exists on one site is a figure nobody has checked, ours included

If you find something wrong, tell us: desk@costcheckfxen.com. Corrections are published, and the bad rows stay in the store flagged — a tick cannot be collected twice, so deleting one destroys the evidence. Who may alter a published figure, and what happens when a broker disputes one, is written out in the editorial policy.