Overview
Aurum AI by Leonid Arkhipov is a gold EA sold separately for MetaTrader 4 and MetaTrader 5. Per the vendor it trades only in the direction of the trend, opens a single trade per signal and sets a stop-loss and take-profit on every trade. Both versions cost $499 on the MQL5 Market, with a free demo and 15 activations. They are among the best-rated gold EAs in both MQL5 top lists. The evidence, however, is a single small signal account, and since version 1.3 a recovery function raises the lot size after losses. Our grade is provisional. The EA is not available at AlgoTradingSpace.
What the Vendor Sells
Per the product page (read 17 Nov 2026):
- Strategy: trend-following on XAUUSD on the M5 chart, “one signal — one trade”. The vendor rules out grid, martingale and “aggressive averaging”. The product name contains “AI”; the description does not explain what part of it is meant to be AI. It speaks of trend analysis and filters.
- Stops: stop-loss and take-profit are active on every trade. With version 1.5 of 28 Mar 2026 the defaults are 1,360 points stop and 650 points target, before that 1,400 and 500 points. Per the release note the EA is meant to trade more often than before.
- Protection: a news filter based on Forex Factory with a pause before and after events, a spread limit and, since version 1.3, a “Safe Recovery function after losing trades”.
- Account: from $100 for 0.01 lots, any reliable broker, standard, ECN and other accounts, VPS optional. The lot size is fixed or calculated automatically from risk; the page names no recommended percentage.
- Bonus: buyers receive a second EA called GoldPrime AI by private message. Several buyers write in the comments that they waited a long time for it; the developer apologised for the delay on 12 Oct 2026.
Marketing, quoted as read on 17 Nov 2026: the description advertises “Discount — 40%” and “Only 3 copies available at the price of $300”. The price shown on the same day was $499 for both versions. It also says “Proven by one full year of real market trading — not just backtests”. The public MQL5 signal has only been running since April 2026; for the time before, the developer points to a cent account outside MQL5 (see below). Whether and how many copies remain at this price we cannot verify. Such an announcement should not drive a purchase decision. Only documented accounts count toward our grade.
Design and Risk
One trade per signal with a fixed stop is one of the lower-risk designs on gold, provided the position size fits. Three points shift the picture for Aurum AI:
1. The stop is much wider than the target. With a 1,360-point stop and a 650-point target, one loss costs more than twice what one win brings. The system therefore needs a high hit rate. As long as it holds, the curve looks calm. A single stop can then cost several weeks of profit. The developer himself confirms a series of losing trades in March 2026 and reset stop and target afterwards.
2. The recovery function raises the lot size after a loss. What exactly it does is not on the product page. On 30 Mar 2026 the developer wrote to a buyer that he recommends 0.01 lots for a $200 account, and that with the recovery function on the lot can rise to 0.02 after a loss. For his earlier cent account he described the same pattern done by hand in July 2025: double the lot after a loss until it is recovered, but no further increase. That is not classic martingale, which doubles again after every loss. It does, however, raise risk exactly in the phase when the account is already down. Because, per the developer, the doubled lot stays until the account has recovered the loss, the decline after a second stop is at least three times that of a single stop, and grows with every further stop. That the function only raises the lot by one step is stated only in a comment, nowhere in the product description. On the signal there have never been two losses in a row so far; how the recovery function behaves in a losing streak is therefore not documented there. Anyone who does not want that should check whether the function can be switched off.
3. Position size and leverage. A buyer reported on 23 Mar 2026 that he lost almost half of his $200 account in a single trade; the developer pointed to the oversized lot of 0.06 and named 0.01 as the ceiling for such an account. With a 1,360-point stop the lot size is the most important dial. The signal runs with 1:1000 leverage; maximum margin usage there was moderate (as of 17 Nov 2026). In the EU retail clients may trade gold at no more than 1:20, major pairs at 1:30. With small lots that is usually enough; with higher risk margin can get tight. How to calculate lot size from stop and account size is shown in the guide position sizing and risk per trade. Why stops, grid and martingale matter so much on gold is explained in gold EA risk: grid, martingale and stop-loss.
The average holding time on the signal is in the range of hours, and standard accounts are allowed per the vendor. We therefore do not list Aurum AI as a scalper. The spread still matters: the developer repeatedly explains missing trades on buyers' accounts with a spread above the configured limit.
The Vendor's Signal
On the product page the vendor links one public MQL5 signal, “Aurum” (2369025). It runs on MetaTrader 5 on his broker's live server, started on 13 April 2026 with $150 starting capital and 1:1000 leverage. There is no separate signal for the MT4 version. It cannot be subscribed to, because MQL5 blocks signals with leverage above 1:500 (as of 17 Nov 2026).
What we read on 17 November 2026, without figures, as in all our reviews:
- The account is clearly in profit. A large part of that came in a single month, July; MQL5 then briefly showed the warning in early August that growth in the last month indicates high risk, and removed it in mid-August. Since then months with gains and losses have alternated.
- The deepest decline is moderate, by balance as well as by equity, and maximum margin usage stayed moderate. Losses have never come two in a row so far. Under our methodology it falls in one of the better bands of the table in the calculation.
- The account trades rarely. Since June MQL5 has repeatedly set and removed the warning that no recent trading activity is visible. In total the history holds only a few dozen trades.
- There were no deposits or withdrawals, which makes the curve easy to read.
Real money or demo: per MQL5 rules only real-money accounts are allowed for signals, MQL5 shows the “Real Account” icon, and the account runs on the broker's live server. Which settings run there is open; whether the recovery function is on is not stated anywhere. We do not run an Aurum AI account in our EA park.
For the time before April 2026 the developer points to a cent account that he presented in the comments in July 2025 and that is said to run on Myfxbook; MQL5 does not admit cent accounts as signals. We have not checked this account. By his own description he doubled the lot by hand after losses there. It does not count toward our grade. How far backtest and live can diverge on gold is shown in gold EA reviews: backtest vs live.
What Buyers Report
The MT5 version stands at 4.81 out of 5 from 47 reviews with 45 comments, the MT4 version at 4.94 out of 5 from 33 reviews with 53 comments (as of 17 Nov 2026). Buyer reviews do not count toward our grade. We relay them because they show what buyers should watch for.
Many reviews praise good backtests, easy setup and quick answers from the developer. Quite a few were written after a few days or the first trade and say little about risk. Critical reports, without judgement on our part:
- Results differ from the signal. One buyer wrote in April and June 2026 that his account took different trades from the signal and lost money over two months; he advises comparing your own trades with the signal. Another asked in June 2026 why the signal opened several trades on one day while his account opened none after a loss. The developer explained a similar difference in April 2026 with different prices per broker: price had turned just short of the target.
- Long periods without a trade. In early 2025 several buyers complained that the EA hardly traded and doubted the backtests; one of them wrote a few weeks later, after an update, that the EA now ran well, and in May 2025 the opposite again. In February 2026 the developer traced missing trades on a buyer's account to a spread that was too high; in September 2026 he explained that the EA deliberately waits for suitable conditions.
- Backtest settings. A buyer reported in June 2026 that in backtests before 2025 stop and target stay at old values whatever one enters. We found no answer from the developer on this.
- Loss with an oversized lot. See above: almost half of a small account in one trade, per the developer due to an oversized lot.
How the Provisional Grade Comes About
The grade follows the same thresholds as for every EA (calculation):
- Risk control: the deepest decline on the signal counts; it falls in one of the better bands. We deduct clearly for the lot increase after losses, which stays until the loss is recovered and has never been tested in a losing streak on the signal, for the stop being more than twice as wide as the target, and for the signal's high leverage.
- Track record: one public account running longer than half a year would earn a good sub-grade under the formula; accounts chosen by the vendor are capped anyway. We lower it clearly: a single small account with few trades, started long after the product went on sale.
- Transparency: public signal, stop and target are in the version history. Deduction because the recovery function is described only in passing, the signal's settings are unknown and the older live evidence is not on MQL5.
- Value for money: $499 per platform falls in the “under $500” band.
- Usability: we have not tested the installation ourselves; the uniform value applies.
- Broker independence: documented at one broker. No scalper deduction, MT4 and MT5 are available.
As with every EA, an AI from a different model family also grades Aurum AI blind; where it differs strongly on a criterion, the mean applies there until a human has re-checked. For Aurum AI that is the case for track record, transparency and value for money. We will review the grade once the signal shows clearly more trades, a second account is added or the vendor describes the recovery function in more detail.
Who It Is (Not) For
Suitable is the design for traders looking for a gold EA with a stop on every trade and few trades a year who deliberately keep the lot size small. Anyone testing it runs the demo in the strategy tester over several periods, clarifies before starting whether the recovery function is on, and compares their own trades with the signal for a few weeks.
Not suitable for anyone already running another gold EA, such as The Gold Reaper or Gold Trade Pro: that is the same bet several times over (see EA portfolio management). Nor for accounts where a single stop at double lot size would hurt, or for anyone extrapolating a month of strong growth into the future. How other gold EAs compare is shown in the overview best gold EA 2026.
Verdict
Per the vendor, Aurum AI is a trend-following gold EA with a stop-loss on every trade and no grid, for $499 per platform. In its favour is a public signal with a moderate decline so far; the good buyer ratings do not count towards the grade. Against it stand a single small account with few trades, a stop more than twice as wide as the target and a recovery function that raises the lot size after a loss. Our grade is therefore provisional. Use it only after your own test and only with money you can afford to lose. We earn nothing from the EA itself; we link the vendor and MQL5 only as sources, never as partner links. Links marked “Ad” on this page (VPS, ATS-listed alternatives) are partner links. Past results are no guarantee of future ones, and trading leveraged products can lead to the loss of the capital invested.
