Leans.ai Review: Remi's Claimed 54.9% On Big Plays, And Why The Published Record Does Not Add Up
Leans.ai sells AI picks from a model called Remi, with a free pick a day and paid tiers. We checked its published record against itself, and against the 52.4% break-even.
Our score
5.4/10Rated 5.4 out of 10
Verdict
Leans.ai is a polite, well-documented picks service with one genuine strength: it grades itself on a published scorecard and explains the unit sizing. That is also where it falls down for us, because the headline record and the sport-by-sport tables do not reconcile, and there is no closing line value figure at all. Use the free daily pick to judge it. We would not pay until the record reconciles and someone independent has audited it.
Best for
Casual bettors who want one free AI pick a day to compare against their own
One free AI pick per day with a free account, so the service can be tried at no cost
The FAQ explains units, vig, lean and the modified Kelly staking in plain English
Tracking is stated to run from September 1, 2022, a longer history than most new tools
Picks are mostly spread, total or moneyline at around -110, with a cap on bad moneyline prices
Covers NFL, NBA, NHL, MLB, college football and college basketball
What held it back
The headline record of 2,670-2,312 does not match the 2-5 unit or 2-10 unit tables
Per-sport units on the 2-10 unit table add to about +923.7, not the stated +563.4
No closing line value figure is published, only win-loss and units
Paid package prices are not shown on the pages we read
All results are the company's own and have not been independently verified
The question we put to Leans.ai was narrow: does a machine-learning model called Remi give you picks that clear the price of the bet, and the price of the subscription, often enough to matter? We used the free daily pick, read the scorecard line by line, and read the FAQ twice. The service is more transparent than most picks sellers. It is also less consistent than it looks.
What you get for nothing
A free account unlocks one AI pick a day. Without one, you see the matchup but the pick, the win probability and the unit size are blurred. That is a fair way to sell: you can watch Remi for weeks before paying.
A Leans.ai college football picks article, opening with the testimonial 'Hard to imagine placing a bet without REMI going forward'
Picks come out on weekdays between 1 and 4 PM Eastern, and weekends between 9 AM and 1 PM. Remi grades every game but releases only a few, typically 2 to 6 a day. The FAQ tells you to treat any game without a pick as a toss-up, which is more candour than most services offer.
The break-even you have to beat
Standard pricing is -110, which means risking $110 to win $100. That needs a win rate of 52.4 percent on spreads just to stand still. Leans.ai states this figure itself, and it is the right yardstick.
The headline claim is a lifetime record of 2,670-2,312 against the spread and +563.4 net units after vig. On the 2-10 unit table the figure is 573-467, a 54.9 percent win rate, with the same +563.4 units. That is 2.5 points above break-even, which would be good if it holds up. Over 1,040 picks, though, 54.9 percent is only about 1.6 standard errors above break-even, so luck remains a live explanation.
The numbers that do not match
Here is the part that stopped us. The header record shows 4,982 decided picks. The 2-10 table shows 1,040. The 2-5 unit table shows 2437-2092, which is 4,529. Those are different samples and the pages do not explain how they relate.
The sport-by-sport table has a bigger problem. NBA +294.2, MLB +257.5, NFL +165.5, CBB +145.5, CFB +111.9 and NHL -50.9 add up to about +923.7 units. The stated total for the same table is +563.4. A gap of roughly 360 units is not rounding. It may be a tracking choice we cannot see, such as picks counted twice, but a reader should not have to guess.
Our view: the service publishes more than most, and that is exactly why the discrepancies are visible. A smaller operator who showed one number would never be caught out. We still cannot treat the record as a verified result.
Remi's FAQ: definitions of units, vig and lean, with a note on the August 2025 upgrade
Units, Kelly and the August 2025 reset
The FAQ is the best page on the site. It defines "units" using a modified Kelly approach, with 1 unit equal to 1 percent of bankroll by default, and notes that quarter-Kelly would be 0.25 percent. A 4-unit play is modeled to win about 54 percent at -110.
In August 2025 the model was upgraded. Legacy v1 and v2 pre-release data, over 1,058 games, was retired, unit scales were normalised to 0 to 10, and the offering was reorganised into three packages. Vault is the elite tier with 6 or more unit plays and strict membership limits. Executive Level covers spreads, totals and props up to 5 units, 10 to 25 plays a week. Intel Level gives 1 to 3 picks a day, including parlays.
Retiring old data is a normal thing for a model to do. For a buyer it means the oldest, weakest track record has been quietly replaced by the newest.
No closing line value, no verdict on skill
The test we care about is closing line value: did the price you took beat the price at kickoff? A bettor who wins 54 percent at -110 might simply have been lucky, while one who consistently beats the close is probably right. Leans.ai publishes no such figure. It reports win-loss and units, which are noisy over hundreds of bets.
The site also leans on volume content. Its AI blog is full of articles on reinforcement learning, live betting and sport guides.
The AI blog: articles on how AI learns and on live betting strategy, beside the Remi mascot
Some of this is useful reading. None of it is evidence that Remi is profitable.
Odds and tools
Odds boards, such as the college basketball page, are a free extra. They list moneylines, spreads and totals for upcoming games and sit beside the picks.
The college basketball odds page: moneylines, spreads and Over/Under totals
Useful, but plenty of free boards do the same.
Who should use it
Use the free tier if you want a second opinion against your own pick and are willing to log the result yourself. If you do, record the price you got and compare it with the closing line. After a couple of hundred picks you will know more than the scorecard tells you.
Do not buy a Vault or Executive package on the strength of the published record. Prices were not shown on the pages we read, which is itself a reason to ask before you commit. We rate Leans.ai 5.4: honest about method, unverified on results. Betting carries real risk of loss, so only stake money you can afford to lose. This review reports what we saw and is not financial advice.
