Gold spread comparison: how much do platforms really differ — and how to measure it yourself

You are watching the screen. The mid is 4,000.095. You hit buy and fill at 4,000.424; flip straight to sell and you would fill at 3,999.766. Same second, same instrument, nothing happened — and your account is already 0.658 USD per ounce lighter. That money never appears in the "commission" column of a statement; it hides in the moment of execution.

That is the spread, and it is the only cost in gold trading that is certainly charged on every trade. Direction can be right or wrong; the spread is never wrong. This article does three things: teaches you to read the spread properly, tells the truth about "how much platforms differ" (sources where I could find them, a plain admission where I could not), and hands you a method for measuring your own spread plus the break-even arithmetic that goes with it.

1. How to read a spread: three units, one meaning

The spread is the gap between bid and ask at the same instant. Using the second this article sampled:

Data: XAUAD live feed /api/config, quote source Swissquote, sampled 2026-10-24 00:32:23 Beijing time (UTC+8), i.e. 2026-10-23 16:32:23 UTC.

Here is something that gets explained wrong again and again: the spread is charged once per round trip, not once on entry and once on exit. Going long you buy at the ask; closing you sell at the bid. The gap is crossed once. So the round-trip spread cost on 1 lot of gold (100 oz) is 0.658 × 100 = 65.8 USD, not 131.6 (for the contract specification and settlement, see what XAUUSD is). What does double is commission — raw-spread accounts often charge "3–4 USD per lot per side", and that really is charged on both sides.

2. Fixed vs variable spread: the difference is who carries the risk

The spread is not a number the platform makes up; it is the price of liquidity. Market makers widen quotes around news releases because for those few seconds nobody wants to take the other side. Whether the platform passes that volatility on to you splits the market into two models.

Which to choose depends on your holding period. If you hold for hours to days, the spread is a rounding error against your P&L and the certainty of a fixed spread is worth more. If you trade 60 seconds to 5 minutes, the spread is the main cost and you must watch the median of a variable spread. That is the starting point of the "three layers of cost" table in gold trading costs fully broken down: trade layer, holding layer, account layer, with the spread only the first.

3. XAUAD's actual spread right now: nine samples

Never mind the marketing page, read the feed. Nine consecutive samples between 00:31:56 and 00:32:20 Beijing time on 24 October 2026, three seconds apart:

Statistics: minimum 0.658, maximum 0.662, median 0.658, mean 0.658. Over the sampling window gold went from 3,999.96 to 3,999.77, a mid fluctuation of about 0.3 USD, and the spread was essentially constant — quotes were stable at the time (Friday New York session, London already closed).

Zooming out: 19 days and 630,000 quotes

Nine samples only describe a quarter of an hour. Digging into this site's own quote database (every bid/ask is timestamped) reveals two completely different regimes:

Source: XAUAD quotes table (quote source Swissquote, every bid/ask/mid recorded tick by tick), split by Beijing calendar day, statistics taken 2026-10-24 00:40 (UTC+8). Spread = ask − bid in USD/oz. The difference in sample counts reflects a change in sampling frequency around 10-18.

That table says more than any marketing copy, and it says three things:

• The mean lies: in the variable regime the mean is 1.489, more than twice the 0.70 median, because extremes like 57.0 drag the average away. Budget your costs off the mean and you overestimate by half; compare platforms off the mean and you are unfair half the time. Look at the median and the 90th percentile.
• Quote structure changes: from 2026-10-18 the feed moved from "wide variation around a 0.70 floor" to "narrow variation around a 0.658 floor", with the median falling from 0.70 to 0.663 and the 90th percentile from 2.90 to 0.70. Same platform, same instrument, completely different cost structure within two weeks — so the sentence "platform X charges Y" only means anything with a date attached.
• The tail never goes away: even in the narrow regime, 0.27% of quotes still exceed 1.0 USD and the maximum was 14.4. The cost in the seconds around a data release is on a different scale entirely, and that is where most stops get taken out.

Spread out across lot sizes, the cost looks like this:

Note the last column: the price move needed to break even is independent of lot size. Both 0.01 and 1 lot need a 0.658 USD move to get even; trading smaller reduces the absolute loss, not the cost ratio. Plenty of people believe "a small position is cheaper to test with", when in fact the cost ratio is identical.

4. "How much do platforms differ": what I checked and what I found

This is the least popular section of the article. Search engines will hand you tables saying "platform X gold spread 0.1", and almost none of them state the capture time or account type, nor whether the figure is the minimum or the average. I am not going to manufacture another one. Here is the raw record of what each broker's site showed on 24 October 2026:

The one industry scale you can actually use: broker XAUUSD variable spreads in normal hours typically sit around 0.2–0.6 USD/oz (this is the rule-of-thumb range used across this site's cost articles, not any platform's official disclosure). The 0.658 measured here is slightly above the top of that range; it was sampled on Friday in the New York session with London already closed, a window where liquidity is starting to thin.

5. Measure your own spread: five steps, half an hour

Step 1: fix three variables

Instrument fixed at XAUUSD, lot size fixed (0.01 is enough — observe only, no adding), account type fixed. Change any one of them and the number can no longer be compared across platforms.

Step 2: sample by session, at least 100 times each

Cover at least three sessions: Asia (08:00–14:00 Beijing), London (15:00–20:00) and the New York overlap (21:00–24:00). Sample every 5–10 seconds; 100 samples takes about 10 minutes. To see which session suits you, compare ranges in the three-session study.

Step 3: record only three fields

Timestamp, bid, ask. Do not record the "spread" conclusion — keep the raw data auditable. The snippet below runs as-is (it uses this site's public endpoint; swap in your own platform's API):

Step 4: compute three numbers, not an average

The median shows the normal case, the 90th percentile shows the worst you will realistically meet, and the maximum shows tail risk. The mean is the least useful of all — a handful of extreme values around news releases pulls it up and makes you misjudge your everyday cost.

Step 5: convert it into "share of target profit"

This is the decision point. With a 0.658 spread against a 2 USD target, cost is 32.9% (live bid/ask is on the today's gold price page); against a 6 USD target it is 11.0%. The same spread means completely different things on different timeframes.

6. How the spread eats short-term win rates: break-even on real ranges

Saying "the spread matters" is useless, so here is the arithmetic. First, two sets of real ranges (from /api/candles, high minus low per bar):

Units: USD/oz. Range = high − low; source XAUAD market feed (Swissquote quotes), captured 2026-10-24 00:33 Beijing time.

Assume take-profit and stop are equidistant (X USD each) and the spread S = 0.658. The hit rate needed to break even is:

To be conservative, recompute with the 0.70 90th-percentile spread from the 10-19 onwards data: the break-even hit rate on a 1-minute median range of 2.17 rises from 65.2% to 66.1%, and on the 4.06 target from 58.1% to 58.6%. A 0.04 USD difference in spread lifts the required hit rate by a full percentage point — which is why "a little bit more" is lethal on short timeframes.

This table explains a common puzzle: why so many 1-minute traders feel they are "right more often than not" and still watch the account shrink. The median 1-minute range is only 2.17 USD; set a 1 USD target and cost is 65.8% of it, so you need an 82.9% hit rate to break even. That is not a skill problem, it is an arithmetic one.

Now the cumulative effect. One standard lot, five round trips a day, 250 trading days a year: 0.658 × 100 × 5 × 250 = 82,250 USD. At 100x leverage and a 4,000 gold price, one lot needs about 4,000 USD of margin — so that spread bill is 20 times the margin. Which is why the life-or-death line for short-term strategies is usually frequency, not signal.

7. Putting the spread into a 60-second to 5-minute decision

No complex model needed — four questions before you order:

1. What is my target profit in dollars? Write the number down first, then compute 0.658 ÷ target. Above 20%, change timeframe or session rather than forcing the trade; run the spread, overnight fee and leverage through the gold P&L calculator to see net profit.
2. Which session is it? First establish whether your quote source is fixed or variable — that decides whether changing session can save you anything. The 583,029 quotes measured in section 3 show the current Swissquote source is fixed: the median spread sits between 0.66 and 0.70 across all 24 hours, and thinning liquidity does not halve it (hour-by-hour detail in sessions and the best trading window). What changing session really buys you is range, not a lower spread; "the overlap has a tighter spread" only becomes true again when the source goes back to variable pricing. Check where you are now on the live session table.
3. Is there data within five minutes? Five minutes either side of NFP, CPI or FOMC, the spread widens markedly (commonly several times normal; the multiple differs by platform and cannot be generalised). In that window stops are more likely to be taken out by the spread than by the market. For reading the macro calendar, see short-term tactics.
4. Am I using a market or a limit order? Market orders add slippage in thin liquidity, so the real cost exceeds the quoted spread; limit orders save cost but may not fill. Both belong in your backtest.

Once you make this routine, you will find far fewer opportunities that actually qualify — and that is itself a return. Get the cost straight first, and direction starts to mean something.

8. Risk warning

9. Frequently asked questions

Q. What counts as a normal gold spread?
A. It depends on the basis. Broker XAUUSD variable spreads in normal hours typically sit in a 0.2–0.6 USD/oz range; this article measured XAUAD at 0.658 USD/oz at 00:32 Beijing time on 24 October 2026, or 0.0164% of the mid. What really matters is not the absolute number but its share of your target profit: above 20%, change session or change target.

Q. Fixed or variable spread — which is better?
A. Neither is absolutely better. A fixed spread does not blow out during news, but you pay a premium for it all the time; a variable spread is cheaper normally, at the cost of multiplying several-fold around major data. If you trade short-term and only in liquid hours, variable usually costs less; if you hold through data and do not want to be swept out, fixed is more restful.

Q. Why is the spread I see higher than advertised?
A. Four common reasons: different account type (standard accounts add a markup, raw-spread accounts charge commission instead); different session (Asia and the close are naturally wider); market makers widening around news; and the platform splitting cost into commission or the fill price. Advertised numbers are usually the "lowest" or "average for one session", not the all-day median.

Q. Is a zero spread real?
A. "Zero spread" usually means a raw-spread account where quotes coincide for a very brief instant, with the real cost taken as commission (commonly 3–4 USD per lot per side, 6–8 USD round trip). On one 100 oz lot that is 0.06–0.08 USD per ounce, which after adding the spread may not beat a spread-inclusive account — you have to add both together.

Q. How do I measure the spread myself?
A. Same instrument (XAUUSD), same lot size, same minute, sample at least 100 times in each of three sessions, recording bid, ask and timestamp, then compute the median and 90th percentile and convert into "spread ÷ target profit". Do not use the "lowest" value as your conclusion, and do not compare prices only around NFP.