Memecoins, and it isn't close. Out of every $100 bet, Polymarket paid back $98.45, the Sweet Bonanza slot pays back $96.50 and pump.fun memecoins paid back $75.36, across 44 million real bets.
Every memecoin timeline is full of 100x screenshots, and nobody posts the other 115,000 buys it took to produce one. So I counted them, rebuilding every trade on pump.fun straight from the Solana chain for the six weeks from 8 August to 18 September 2026, which came to 44 million bets placed by 6.7 million wallets.
A bet here is one wallet buying into one coin, counted from its first buy until its last sell (or until the end of the six weeks, valuing whatever bag was left). Then I did the same for Polymarket, with every fill on Polygon over the same six weeks: 21 million bets from 328,000 wallets.
The online casino side needs no data at all, because games like roulette, Plinko and Sweet Bonanza run on fixed odds that are known to the decimal. That puts all three on the same scale, and the full paper with every test is coming soon.
**Before anyone says "skill issue"
I took out everyone who isn't a regular trader: the wallet that created the coin, wallets that bought in the same block as the launch, bots that buy and sell inside one transaction and any wallet trading more than 200 times an hour, which leaves people pressing buttons.
The rules were written down and timestamped on the Bitcoin blockchain before I computed a single profit or loss, so nothing here was tuned after seeing the answer. After that I rebuilt the whole memecoin side from an independent second data source and got the same number.
How much each one keeps
That $75.36 makes memecoins worse than every casino game in the comparison. American roulette, the worst bet on a casino floor, gives back $94.74 and European roulette $97.30, while craps, baccarat and Plinko all return close to $99.
Polymarket's $98.45 sits between the two roulettes and those three games, which puts it right in the middle of the casino games.
Over the six weeks, regular memecoin traders put in 46.4 million SOL and took out 34.9 million, a net loss of 11.4 million SOL. At the window's average SOL price of $93 that is a little over $1 billion, whereas Polymarket's regular traders lost $28M on $1.8B of bets over the same period.
Per wallet, the average memecoin trader put in 6.9 SOL (about $640) and lost 1.7 SOL of it (about $160). The average Polymarket wallet bet far more, about $5,500, and lost $86.
The odds of hitting a 100x
The whole reason to buy a stupid memecoin is the shot at a 100x or more, so these are the real odds, per buy:
- 2x or better: 1 in 47
- 10x or better: 1 in 2,519
- 100x or better: 1 in 115,484
- 1,000x or better: 1 in 3.4 million
Three out of four buys lose money, and the median buy loses 32% of what went in. Of the roughly 1.1 million coins launched in the window, only about 3 in 100 ever filled their bonding curve.
The suprising comparison is with Polymarket. A Polymarket bet returns 100x or more once every 2,685 bets, about 43 times as often as a memecoin buy, because buying "yes" at a cent on something that then happens pays exactly what it says.
A Plinko ball at an online casino does it once every 1,928 drops. If what you want is a lottery ticket, memecoins are a bad lottery, with big multiples rarer than on Polymarket and rarer than on the casino game built to sell big multiples.
Why memecoins lose differently from a slot
Slot machines take your money in small steady bites and pay some of it back as rare jackpots, which is the whole design. Memecoins are missing the jackpot half: counting every buy as the same size, hits of 100x or better make up 0.36% of everything paid back, against 9.5% on a Plinko board.
The losses come instead from millions of ordinary buys that each lose a large share of the stake. Bet size makes it worse at both ends, with bets under 0.1 SOL getting back 72 per 100 and bets of 1 to 10 SOL getting back 88, since every transaction carries a fixed network cost that bites harder into a small bet. At the other end of the scale, bets over 10 SOL did worst of all at 57 per 100.
What happened to the wallets
Single bets are one thing, but what matters is whether a wallet is up at the end. After six weeks:
- 14.1% of memecoin wallets ended ahead, against 39.0% on Polymarket.
- 49.1% of memecoin wallets lost 90% or more of everything they put in, against 11.7% on Polymarket.
- 1.4% of memecoin wallets doubled their money, against 2.5% on Polymarket.
Three in four memecoin wallets placed exactly one bet in the six weeks, and only 14.7% of those one-shot wallets came out ahead. To put these numbers next to a casino, I took every wallet's exact sequence of bet sizes and replayed it at casino games 200 times each.
On the Sweet Bonanza slot, 19.7% of those replays end ahead, and on European roulette 48.3% do. A memecoin trader placing the same bets on a slot machine would have finished ahead more often than they did on pump.fun.
Who is taking the money
The money doesn't disappear. It moves, and the groups I removed from the regular pile show where:
- Token creators got back 156 SOL for every 100 they put in, a net gain of 1.52 million SOL.
- Wallets that bought in the launch block got back 149, a net gain of 1.49 million SOL.
- Bots mostly lost too, with wallets trading over 200 times an hour getting back 85 and bots that buy and sell inside one transaction getting back 76.
Creators and launch-block snipers alone walked away with about 3 million SOL, close to $280M at the window's average price. Both groups are in before anyone else can buy, so the regular traders who arrive later are the people they sell to.
So if you are not the exception, then you are the product.
Is anyone actually good at this?
A few are. Of the wallets with at least ten bets, 10.6% did better than luck would explain, against the 5% that chance alone produces, so a skilled minority exists on pump.fun.
They don't trade like a person, though. Ranking active wallets on the first three weeks, the top 10% bought a median of 2 minutes after the coin was created, held for a median of 62 seconds and put in about 0.81 SOL a bet, which points to scripts rather than people reading charts.
Even they couldn't keep winning. Over the next three weeks that same top 10% got back 98 per 100, far better than everyone else's 72 and still a loss.
Polymarket isn't free money either
Polymarket shows no skill at all. Only 2.2% of wallets with at least ten bets beat their own luck benchmark (fewer than the 5% chance alone would give), and the first three weeks' top 10% did slightly worse than everyone else over the next three, at 97.6 per 100 against 98.3.
It behaves like a casino in another way as well. Of wallets that placed a single bet, 47% ended ahead, which falls to 33% for wallets with 10 to 99 bets and 21% for wallets with 100 or more, as a small edge against the player hides inside a few bets and shows up reliably over many.
Most regular traders also sit in the worst corner of the site. Short-term crypto up-or-down markets hold 61% of their bets and pay back 97.6 per 100, against 99.7 on politics and 99.6 on sports.
The usual objections
"You counted bots as retail." Tightening the bot rule from 200 trades an hour to 50 makes regular traders look worse, at 68.6 per 100, because it removes fast wallets that were doing relatively well. Loosening it to 1,000 an hour gives 75.4, so no version of the filter gets anywhere near a casino.
"I don't ape the launch, I buy later." Bets opened more than an hour after launch got back 81 per 100. That is better than the average and still well below American roulette.
"What about bags people are still holding?" Every bag left at the end was valued by selling it into the coin's pool at the final price, slippage included, while every fee and the network cost of every transaction is counted on the way in and out.
"Polymarket just had a good six weeks." A separate run on Polymarket from February to May 2026 gave 99.4 per 100 with 41% of wallets ahead, close to these six weeks.
So, which is worse?
If you are going to gamble, it helps to know the price. On pump.fun it was about 25 cents of every dollar, against 3.5 cents on a slot machine and a cent and a half on Polymarket.
The full paper is coming soon, with every test, the pre-registration and a third venue: Hyperliquid perps.
