Gold’s record highs:how 5,586 happened, and what 26 years of data say

On 29 January 2026 the front-month COMEX gold contract touched USD 5,586.20 intraday and settled at USD 5,318.40. Both were all-time highs. Nine months later, on 23 October 2026, the same contract closed at USD 4,007.90 — still 24.64% below that peak.

“Gold hits a record high” is the most overused sentence of this cycle. Almost nobody unpacks it: which record? Intraday or the close? Futures or spot? What normally happens next? And where does the 5,586-to-4,008 drawdown rank in the last 26 years?

This article answers with 6,563 daily bars (30 August 2000 to 23 October 2026) and 386 record-close days, and every figure can be reproduced from the sources listed below. For anyone holding a non-dollar account there is one more layer — you see a dollar gold price while your account may be in dollars, yuan or Singapore dollars — and the FX overlay is dealt with separately at the end.

1. Define “record high” first: three measures, USD 268 apart

On the same day for the same instrument, “a record” can mean three completely different numbers. Confuse them and every subsequent “how far has it fallen” calculation is wrong.

Intraday and close differ by USD 267.80, close to 5%. If a headline says gold reached 5,586 while your account shows 5,318, neither is wrong — they are different measures. Our review of an 8.2% monthly decline uses the Swissquote spot measure, USD 4,331.53 (2026-09-25 19:10 Beijing), which is the high of that 30-day window and not the all-time high. Do not cite the two interchangeably.

Why be pedantic? Because a basis exists between futures and spot and it moves. The same-moment basis above is +10.01 (0.25%), inside the normal range — but during a squeeze like January 2026 the basis widens sharply, and using a futures “record” to describe the P&L in a spot account produces an error an order of magnitude larger than the spread.

2. Twenty-six years in one table: the gold price by year (open, high, low, close)

The table below shows the opening price on the first trading day and the close on the last trading day of each year for the GC=F front contract, with the intraday high and low for the year. It reveals something counter-intuitive: gold does not rise slowly — it sits flat for years and then explodes.

Source: Yahoo Finance GC=F daily bars, aggregated by calendar year, pulled 2026-10-25. 2026 is a partial year, closing price as of 2026-10-23. Cumulative gain over 26 years is 14.63×, or 10.81% annualised (2000-08-30 to 2026-10-23, 365.25 days per year).

Read down the column and you find the two negative years, 2013 and 2015 — the two-and-a-half-year bear market that followed the 2011 top. You also find the +62.6% of 2025, the only year in 26 above 50%. Put differently, thirteen of the twenty-six years gained more than 20%, and five lost money: “gold rises over the long run” and “you made money this year” are two different statements.

3. Only 386 record days: 5.9% of all sessions

Of 6,563 trading days, just 386 set a record close — 5.9%. That ratio tells you records are rare, but their rhythm is more interesting:

Broken out by year it becomes clearer still — records are not evenly spread, they arrive in bursts concentrated in a few years:

The longest drought: the USD 1,888.70 of 22 August 2011 was not exceeded until USD 2,089.70 on 1 December 2023 — 4,484 days, 12.3 years. Gold did not fall the whole time (it gained 7.0% in 2016 and 18.6% in 2019), it simply never once set a new high. For anyone holding longs, that is the most exhausting kind of market: the account moves, but never gets back to the “all-time high” you heard about when you entered.

4. How 5,586 was built: ten records in thirteen sessions

All ten record days of 2026 fall inside the thirteen sessions from 12 January to 29 January. Price ran from 4,614.70 to 5,318.40, a 15.2% gain in thirteen sessions. That is not a slow climb; it is a textbook accelerating blow-off.

What happened the next day deserves its own note: 30 January 2026 fell 10.78%, from 5,318.40 to 4,745.10. That is the biggest one-day drop since 2000, deeper than the famous April 2013 “gold crash” (−9.35%).

To rule out a single bad data source, the same event was cross-checked against the gold ETF GLD: it closed at 495.90 on 2026-01-29 and 444.95 on 2026-01-30, a one-day −10.28%. Two entirely independent instruments gave the same direction and the same magnitude, which means this is not a GC=F data error but a real market move.

What that record means for short-term traders: risk is asymmetric on the record day and the day after. The record day is when the buying is most crowded and when liquidity is thinnest and slippage worst.

5. Five major tops, five very different repair times

Over 26 years there are five occasions when price fell more than 20% from an all-time high. Set side by side, they show the truest thing about a top: it is not how far it falls, but how long it takes to get back.

Method: all figures are GC=F closes, with tops and bottoms taken as extremes of the close series; repair days = calendar days from the bottom to the day the previous top was regained, counted on the source’s actual trading calendar. The 2026 low uses the close of 3,992.10 (2026-07-16); the intraday low was 3,955.40 (2026-06-30), USD 36.70 lower.

Three usable conclusions:

• A 20%–25% drawdown is normal, not exceptional. Three of the five sit in that band, and the 2026 episode (−24.94%) ranks third.
• Repair time ranges from 302 to 1,680 days — a 5.5× spread. Both were “a twenty-percent fall”: 2008 recovered in ten months, 2011 needed four years and seven months.
• Today’s 4,007.90 is only +1.33% above the low of 3,955.40. The 23 October close is parked 1.3% above the July low, which is a position where “the fall is over” has not been proven. The arithmetic of being underwater is worked through in stuck in a gold loss: averaging down, stops and the maths of breaking even; the one point to stress here is that after −24.94% you need +33.2% to get back (1 ÷ 0.7506 − 1).

6. Twenty days after a record: forward statistics on 386 samples

“Should I chase a new high?” is answered more directly by forward-testing 386 record days than by any opinion. Method: take each record day’s close as the starting point, then measure returns 5 and 20 sessions later, plus the deepest drawdown inside those 20 days.

The reading is this: after a record, price probably keeps rising (median +1.77%), but nearly four times in ten it loses money, and the worst tenth loses more than 4.77% inside 20 days. It is a positive expectancy, not a low-risk trade. Treating it as “buy with your eyes closed” is treating a 37.8% probability as zero.

Compare it with the 2026 sample in section 4: buying the first record on 12 January was worth 15% by 29 January; buying the record on 29 January lost 10.78% the next day. The same “new high” signal has opposite odds at the start of a trend and at the blow-off. The tool for telling them apart is not the record itself but the density of records: ten records in thirteen sessions is acceleration, not initiation.

7. Inflation-adjusted: 2011’s 1,888 is about 2,809 in today’s dollars

The thing most often left out of any “record high” discussion is inflation. A nominal record and a purchasing-power record are two different events. Deflating with US CPI-U (FRED CPIAUCSL):

Deflator = 336.298 ÷ 226.106 = 1.4873. CPI source: FRED CPIAUCSL (monthly, not seasonally adjusted), 2026-09 being the latest print, pulled 2026-10-25. The table adjusts only for purchasing power; it makes no allowance for tax, storage or opportunity cost.

Three sentences follow from the conversion:

• In nominal terms the 2026 high (5,318.40) is 2.82× the 2011 high (1,888.70);
• After inflation it is only 1.89× — half of the gain is the dollar losing purchasing power;
• But even measured in 2011 money, today’s 4,007.90 is worth 2,694.66 of then-dollars, still 42.7% above the 2011 record.

That is the real boundary of “gold preserves value”: it preserves long-run purchasing power, not every year’s. 1.89× over fifteen years is about 4.3% annualised — ahead of inflation, behind equities in 2024–2025. For the deeper version of this argument, see does gold really hold its value: 50 years of data, three methods.

8. Putting records and drawdowns to work on a 60-second to 5-minute order

Everything so far is daily-bar material. When you actually place an order you face 1-minute and 5-minute charts and a spread that does not shrink. Start with our own 30-day measurement:

Put the spread next to the median range and the cost share falls out: on a 1-minute trade the spread eats 68.2% of the median range; on 5 minutes, 30.4%. One step further — how many bars have a range big enough to cover the spread at all:

“Range > spread” does not mean “you make money”; it is only a necessary condition. A bar with a big range means there is something to win if you guess the direction right. Only 61.3% of 1-minute bars clear that bar, which means in nearly four minutes out of ten the move is too small to cover the spread no matter which way you call it. That is the hardest part of short-term trading — not seeing it wrong, but seeing it right and still not being paid.

The matching break-even win rates (odds from the current XAUAD configuration, /api/config: 60 s 0.68, 300 s 0.80, 900 s 0.85, 3600 s 0.90):

Windows beyond 60 seconds were not measured separately because the range over 900 or 3,600 seconds is set by the intraday trend and is no longer “the range of a single bar”. The numbers above give three rules directly:

• While the drawdown is unrepaired, do not counter-trend scalp a 60-second on the dip. We are −24.94% from the high and 1.33% off the low, direction unproven; a 60-second needs 59.52% to break even while the spread eats 68.2% of the median range. Going against the trend on 60 seconds means paying the highest cost to bet on the least certain direction.
• Do not chase a record day on 60 seconds. Ranges expand on the day (the 29 January intraday range was USD 488.70), which looks profitable, but the same minute also produces the most two-way false breakouts. If you want in, push the expiry to 300 seconds or more and give the trend time to complete.
• Use range ÷ spread as the switch, not your feelings. In the minutes where the 1-minute range falls below 0.662 (about 38.7% of them), any 60-second trade is a negative-expectancy structure. Simply do not take it.

The largest 1-minute bar in the window was USD 10.69 at 2026-10-22 05:38 UTC, 16.2× the current spread. Such minutes exist, but they are unpredictable and non-contiguous — building a strategy on them is buying a lottery ticket. For how ranges are distributed, see how to measure gold volatility: ATR, session ranges and 60-second odds; for how to match expiry to conditions, see short-term gold tactics: the complete 60-second to 5-minute playbook.

9. One more layer for non-dollar accounts: FX

Every number above is a dollar gold price. If you fund in yuan, Singapore dollars or ringgit, your P&L carries an extra FX layer: dollar gold up 1% with your currency up 1% against the dollar leaves you roughly flat. In the 2026 episode dollar gold fell from 5,318 to 4,008 (−24.64%); if your currency depreciated over the same period your real loss was smaller, and if it appreciated the loss was larger. The conversion formulas and true costs are set out in USD gold vs CNY gold: conversion formulas and the arbitrage reality, and today’s readings for both are updated every five minutes on the gold price page.

10. FAQ

Q. What is the actual all-time high for gold?
A. On the COMEX front contract (GC=F) the record close is USD 5,318.40 and the intraday record is USD 5,586.20, both on 2026-01-29. Note that these are futures measures; Swissquote spot was about 10 dollars lower at the same moment.

Q. How far is the price from the record now?
A. The 23 October 2026 close was USD 4,007.90, which is −24.64% from the 5,318.40 high, and only +1.33% above the 2026 low of 3,955.40. Getting back to the high requires a 32.7% gain.

Q. Can you still chase gold after a new high?
A. Forward-testing 386 record days gives a median 20-session return of +1.77%, but 37.8% of cases were negative and the worst tenth lost more than 4.77%. Positive expectancy, not low risk — size for a 37.8% failure rate.

Q. Why were there no new highs for twelve years from 2011 to 2023?
A. After the 1,888.70 top on 22 August 2011 the maximum drawdown was −44.36%; the bottom of 1,050.80 only arrived on 17 December 2015, and another 1,680 days passed before the old high was regained. Between 2012 and 2019 there was not a single record day for eight straight years.

Q. Is a record day a good time for a 60-second contract?
A. No. Over our 30-day measurement the median 1-minute range is only USD 0.97 while the 0.662 spread already accounts for 68.2% of it, so a 60-second needs a 59.52% win rate just to break even — and record days produce denser two-way false breakouts. If you want to participate, extend the expiry to 300 seconds or more.