Memecoin trading, honestly: how it works, what it costs, and the real odds

Posted by the operator of RebateMax

I earn a referral commission from BingX and LBank when you register through my links, and both exchanges list meme coins. So I have a stake in one part of this post: the section that compares trading on-chain with trading on an exchange. I have put the costs and the odds before that section, not after it. I have no relationship with pump.fun, with any trading terminal, or with any trader named below, and none of them has seen or endorsed this post. Every third-party figure was checked against its source on 21 September 2026 and is linked at the end. This market changes monthly, so every number carries its date.

The short version

• Almost every coin dies. In a Carnegie Mellon and EPFL study of pump.fun's first two years, 1.02% of 15.2 million coins ever left the launch pool. Solidus Labs found 98.6% of 7 million tokens fell below $1,000 of liquidity.
• Most wallets make or lose pocket change. CoinGecko's April 2026 count put 65% of 3.1 million wallets at a realized profit between $1 and $500, and 5.4% above $1,000. It counts only coins people sold, not the dead ones they still hold.
• The cost is large and fixed. pump.fun charges 1.25% per trade on its launch pool and a terminal charges about 1% on top. Buying and selling once costs about 4.4% before slippage.
• The famous traders are paid by your trading, not only by theirs. Axiom, the largest terminal, pays a referrer 30% of its fee on every trade a referred user makes. The leaderboard that ranks these traders is owned by pump.fun.
• An exchange is a different product, not a cheaper version of the same one. It costs a fraction as much per trade, and it only lists a coin after the coin has survived. You cannot buy a one-hour-old coin there.

What a memecoin is

A memecoin is a crypto token whose value comes from attention rather than from a product, a cash flow or a protocol. It has a name, a picture and a community, and its price is whatever the next buyer pays. That is not an insult. It is the definition, and it explains everything that follows: a memecoin's price is a measure of how many people are looking at it right now.

Some are years old and trade on every large exchange. Most are minutes old and will be gone by tomorrow. The rest of this post is mostly about the second kind, because that is where the phrase "memecoin trading" now lives.

How pump.fun works

Almost all new memecoins are launched on Solana, and most of those on pump.fun. The CMU and EPFL researchers counted 15,245,966 coins created there between 14 January 2024 and 14 January 2026, a median of 17,926 new coins a day, with a record of 71,735 in one day. Their paper is a preprint from September 2026 and has not been peer reviewed yet.

The mechanism has three stages.

Creation. Anyone types a name and a ticker and uploads an image. pump.fun mints one billion tokens and places them in a bonding curve, which is a pool that sets the price by formula. Every purchase pushes the price up along the curve and every sale pushes it down. There is no order book and no seller on the other side. You are trading against the formula.

Graduation. When buyers have emptied the curve, which for coins priced in SOL happens at roughly 85 SOL paid in, the coin graduates. Since 20 March 2025 graduated coins move to PumpSwap, pump.fun's own exchange. Before that date they moved to Raydium. In the study's data, 1.02% of coins got this far.

After graduation. The coin now trades in an ordinary liquidity pool, where the price is set by what the pool holds. pump.fun's co-founder has said publicly that most gains are made after a coin graduates. The remark was made in January 2025 while disputing a profitability statistic, so treat it as the platform's view, not a neutral finding.

Here is what pump.fun charges, from its own fee page as last updated on 20 May 2026.

Per trade, charged on the buy and again on the sell. The creator's share on PumpSwap slides down as the coin's market cap rises. A graduation costs the pool 0.015 SOL.

The same study measured what this open door lets in. At least 1.5 million coins, more than 10% of all coins, were copies of an existing coin. 23.5% of coins were created shortly after a post on Twitter or Truth Social. The most active 1% of creator groups, when wallets were grouped by who funded them, made between 53% and 58.6% of all coins depending on how far the funding was traced. In the study's samples, 17% of trades by count were wash trades, meaning the same party on both sides. By volume those trades were only 1.32%, so they inflate a coin's activity more than its money. The researchers also found tools sold to people with no technical skill for doing exactly this.

What a trading terminal does

Few people trade on pump.fun's own page. They use a terminal: a faster screen that shows new coins, coins about to graduate and coins that just did, with one-click buying and labels on the wallets of known traders. Axiom and GMGN are the largest. Photon and BullX are older names.

The terminals charge about 1% per trade on top of pump.fun's fee. Axiom's own documentation lists its net rate from 0.95% at the entry tier down to 0.75% at the top tier. GMGN, BullX and Photon each document 1%. According to DefiLlama, users had paid Axiom $799.5 million in fees in total and $50.0 million in the 30 days to 19 September 2026.

Two things about terminals matter more than the fee. The first is custody. A web terminal asks your own wallet to sign each trade. A Telegram trading bot usually creates a wallet for you and holds its private key, which means the bot's operator, or anyone who breaches it, can move your funds. Only keep in a bot what you could lose tonight. The second is the referral scheme, which I will come to under the famous traders, because it explains their incentives better than anything they post.

What one round trip really costs

This is my own arithmetic on the published rates above. It is a floor, not an estimate of what you will pay, because it leaves out slippage, Solana priority fees and any tip paid to land the trade quickly. On a coin that is minutes old, slippage alone can be larger than everything in this table.

Fees only, each side charged once. On-chain rows use pump.fun's published 1.25% and 0.30% plus Axiom's published entry rate. BingX rows use its standard 0.1% spot rate, halved by the referral discount. Slippage, network fees, exchange spreads and withdrawal fees are excluded from every row.

Two honest notes about that table. I do not have a verified LBank spot figure for this post, so I have not printed one; the rates I can state for LBank are for perpetual futures, which is a different market. And the exchange rows are not the same trade as the on-chain rows. They are cheaper because they are a different product, which is the subject of the section after next.

At thirty round trips a week of $500 each, the first row is $660 a week in fees before a single coin has moved. The fee calculator runs the exchange side of that on your own numbers.

What the odds really are

Three large studies are quoted constantly, usually one at a time and usually to prove whatever the writer already believed. Put side by side they do not contradict each other. They measure different things.

Each figure is the publisher's own, with the caveat in the publisher's own terms. "Realized" means the profit or loss on coins that were actually sold.

Read together they say this. Nearly every coin dies. Most wallets that sell come out a few dollars ahead or behind, because most people buy small and sell fast, and a loss only counts once you sell. The dead coins still sitting in wallets are invisible to the two profit studies. A very small number of wallets make very large sums, and the studies cannot tell you how many of those are bots or the coins' own creators.

One more study is worth knowing if your plan is to copy a successful wallet. A paper accepted at the 2026 Web Conference proves, from the bonding curve's own formula, that a copier "strictly overpays on every replicated buy," because the copied wallet's purchase has already moved the price. In the authors' tests on 6,000 coins, standard methods that correctly identified profitable wallets still produced negative returns for the copier. Their own, more elaborate system picked wallets that averaged 14% and estimated the copier's return at 3% per coin. That is the best case in the paper, and it is before anyone reacts to being copied.

Who the famous traders are, and how the money works

You will see the same names on every terminal and every timeline. Traders in this scene are called KOLs, for key opinion leaders. I am naming a few because you will meet them anyway, and I am stating only what each has said or what a named source has reported. I could not check follower counts first-hand, so I have not printed any.

• Murad Mahmudov gave the "memecoin supercycle" talk at Token2049 in September 2024. Cointelegraph reported in October 2024 that the investigator ZachXBT had linked eleven wallets to Mahmudov, holding large positions in the coins discussed in that talk. No wrongdoing was found or charged. It is a useful example of the general point: a public figure's public wallet is not their only wallet.
• Orangie is a former professional Fortnite player who streams and films memecoin trading. By Orangie's own account, posted on X, the starting stake was about $10,000, it grew to about $1 million in the NFT cycle, and it was held to near zero in late 2022 before the rebuild.
• Cented streams trading live. A wallet that Kolscan labels as Cented's was seventh on Kolscan's daily board on 21 September 2026, up 195.42 SOL for the day. That is one labelled wallet on one day, not a track record.
• Ansem, Cupsey and Threadguy are names you will also see constantly. I have no primary source for anything about them beyond their own posts, so I will leave it at the names.

Now the part that matters. There are five ways a well-known memecoin trader gets paid, and only one of them is trading well.

Trading. Real for a few, and hard for an outsider to verify, for the reason above.

Referral fees. Axiom's documentation states that it pays a referrer 30% of its net fee on the trades of everyone they refer, plus 3% on the people those people refer and 2% on the level below that. A trader with a large audience is paid on every trade their audience makes, whether it wins or loses. I know this model well because a version of it is my own business, which is why I disclose it at the top of every post.

Paid promotion. On 1 September 2025 ZachXBT published a spreadsheet from a single token's marketing campaign. The Block reported that the documents appear to show more than 200 influencers were approached, about 160 accepted payment, and fewer than five labelled their posts as advertising. The Block did not name them and neither will I. Regulators treat this seriously when they can reach it: in 2022 the US Securities and Exchange Commission fined Kim Kardashian $1.26 million for promoting a token without disclosing a $250,000 payment for the post. The settlement was reached without admitting or denying the findings.

Content. Streams, videos, sponsorships and paid groups.

Attention itself. When a widely followed wallet buys, copiers follow, and the price rises on their buying. The copy-trading paper above shows why the copier is structurally the one who pays for that. I am describing the mechanism, not accusing anyone of using it.

And one fact about the scoreboard. Kolscan, the leaderboard most people use to rank these traders, was bought by pump.fun in July 2025, as CoinDesk and The Block reported. The house owns the leaderboard. DL News reported that a group of connected wallets made more than $3 million buying a Kolscan-related token before the acquisition was announced, and named no one. A leaderboard ranks wallets that someone has labelled. Labels can be wrong or out of date, and nothing requires a trader to use only the wallet you can see.

On-chain or on an exchange

Most guides are written by a terminal or by an exchange, and each tells you its own route is the smart one. They are different products and the choice is a real trade-off.

A comparison of structure, not a recommendation. Custody risk on an exchange is real, and I have written about it for BingX specifically.

If what you want is a lottery ticket on something an hour old, only the on-chain route sells it, and the price of the ticket is the first table in this post. If what you want is exposure to a meme coin that already exists and already has a market, the exchange route costs a small fraction as much per trade, as the cost table shows, and removes most of the clone problem.

What we measured ourselves

I run a page that tracks every meme coin listed on BingX and LBank: the Memecoin Desk. Building it taught me more about this market than reading about it did, so here is what happened, including the parts that went wrong.

The first version looked coins up by their ticker. All five coins it measured were the wrong coin. It reported a market cap of $403 million for a coin worth $2.1 million, because another token had the same name and more liquidity. A ticker identifies nothing. The contract address is the only identity a coin has.

The second version checked the contract. It then found three tokens on Solana named PEPE, each priced within 3% of the real PEPE's price, each with a pool whose stated liquidity was almost equal to its whole market cap. They were built to pass exactly the check I had written. A real coin's pools hold a small fraction of its market cap. A clone's pool can hold all of it, because the clone's creator owns all of it.

The third problem was a real coin with a fake price. One pool for a genuine, exchange-listed token quoted $21.47 while both exchanges traded it at $0.0044, which put its market cap at $21 trillion. The page now measures a coin only from pools that agree with the exchange's own price, and prints nothing when none do.

As of 21 September 2026 the page carries 45 coins with a verified contract and 6 exchange listings with no numbers at all, each with the reason stated. These are the seven meme coins my research lane had written up before the page existed, measured from the day each was covered.

Market cap from on-chain pools, by verified contract. Seven coins over six weeks is not a sample and proves nothing about what comes next. These coins also passed a published checklist before they were covered, so they are not typical of the market. They were measured, not endorsed.

The other 38 coins on the page have been measured since 20 September 2026. The page will not show a change for any of them until it has a week of history, because a one-day change is noise.

If you still want to try it

Treat the first month as tuition and size it that way.

1. Decide the amount you are fully prepared to lose, then the size of each trade. A common rule is 2% to 5% of that amount per trade, and a hard stop for the day. Most accounts are lost in one evening of trying to win a loss back.
2. Use a new wallet that holds only this money. Write the seed phrase on paper. Never type it into a website, a form or a chat, and never send it to anyone claiming to be support.
3. Fund it by withdrawing from an exchange, with a small test transfer first. Keep a little SOL aside for network fees.
4. Pick one terminal and reach it only from a bookmark of its official address. Fake terminal sites and fake support accounts exist to drain wallets.
5. Never run a trading bot from GitHub with a real key on your main computer. In July 2025 SlowMist documented a popular "solana-pumpfun-bot" repository, with inflated stars, whose dependencies searched the machine for private keys and uploaded them. In March 2026 StepSecurity documented a working trading bot hosted in a hijacked, verified GitHub organization that stole key files and opened a remote-access port. In January 2025 Socket found packages that emptied 98% of a victim's wallet. A bot that really works is not evidence that it is safe.
6. Check the contract address, not the name, against the coin's own official channel, before every first purchase.
7. Read the coin before you buy it. Who holds the top ten positions, how much the creator still holds, how many wallets bought in the launch block, and whether the chart climbs in a staircase too smooth to be people.
8. Start with graduated coins, not brand-new ones. The launch pool is where the bots are fastest and the fee is highest.
9. Decide the exit when you enter. Taking your stake out when a position doubles is a common rule. So is never adding to a loser.
10. Keep a record of every trade from the first day. In many countries each swap is a taxable event, and a busy month produces hundreds. I am not a tax adviser. Ask one before the numbers get large, not after.

Questions people ask

Can you make money trading memecoins? Some wallets do. In CoinGecko's April 2026 data 5.37% of wallets had realized more than $1,000, and in the January 2025 Dune data 0.4% had realized more than $10,000. Neither figure counts dead coins still being held.

What percentage of memecoins fail? In the CMU and EPFL study, 98.98% of pump.fun coins never graduated. Solidus Labs found 98.6% of traded tokens fell below $1,000 of liquidity.

What happens when a pump.fun coin graduates? Its bonding curve is complete, at roughly 85 SOL for SOL-priced coins, and it moves to a normal liquidity pool on PumpSwap, where the fee falls as the coin grows.

How much does pump.fun charge? 1.25% per trade on the bonding curve and between 0.30% and 1.25% on PumpSwap, as of its fee page dated 20 May 2026. A terminal adds about 1%.

Is copying a successful trader's wallet profitable? A 2026 study proves the copier pays a higher price than the wallet being copied on every buy, and found negative copier returns even when the profitable wallets were identified correctly.

Is it safer to buy memecoins on an exchange? It removes most clone and signing risk and costs far less per trade. It adds the risk of the exchange itself, and you can only buy coins that were already listed.

Who owns Kolscan? pump.fun, since July 2025.

What to do with this

1. Work out your own round-trip cost before your first trade, using the first table, and multiply it by how often you expect to trade in a week.
2. Read the odds table again and decide which of the three studies describes the trader you expect to be.
3. If you follow a well-known trader, find out which terminal they link to and remember that they are paid when you trade, not when you win.
4. If a coin you want already trades on an exchange, compare the two routes on cost before choosing. On BingX spot, registering through my link halves the standard rate. It does nothing about the coin going to zero, and I will not pretend otherwise.
5. Check every figure here against its source below. This post is dated, and this market will have moved by the time you read it.

Sources

• Szwajcok, Tsuchiya, Liu, Christin, Soska and Payer, "Meme Coin Factories: Uncovering Large-Scale Manipulations on pump.fun" (preprint, submitted 9 September 2026; data 14 January 2024 to 14 January 2026): arxiv.org/abs/2609.10246. Source for the coin counts, the graduation rate, copycats, creator groups, post-triggered launches and wash trades.
• Luo, Feng, Xu and Liu, "Resisting Manipulative Bots in Meme Coin Copy Trading" (The Web Conference 2026): arxiv.org/abs/2601.08641. Source for the copier's overpayment and the copier return figures.
• pump.fun, fee documentation (last updated 20 May 2026): pump.fun/docs/fees. Source for every pump.fun and PumpSwap fee.
• pump.fun, public program documentation: github.com/pump-fun/pump-public-docs. Source for the one billion supply and the bonding curve.
• The Block, on the move from Raydium to PumpSwap (20 March 2025): theblock.co/post/347360.
• Axiom, fee tiers and referral program (read 21 September 2026): docs.axiom.trade/getting-started/fees/axiom-fees and docs.axiom.trade/getting-started/referral-program.
• GMGN, BullX and Photon fee pages: docs.gmgn.ai/index/gmgn-fees-settings, bullx.gitbook.io/bullx-neo-docs/fees-and-gas, pies-organization.gitbook.io/photon-trading.
• DefiLlama, Axiom fees (19 September 2026): defillama.com/protocol/axiom.
• Decrypt, on the Dune profitability dashboard and the co-founder's reply (10 January 2025): decrypt.co/300403/pump-fun-traders-millionaires.
• CoinGecko Research, "Pump.fun traders are making a comeback" (7 May 2026): coingecko.com/research/publications/pump-fun-traders-are-making-a-comeback.
• Solidus Labs, Solana rug pulls and pump-and-dumps report (May 2025): soliduslabs.com/reports/solana-rug-pulls-pump-dumps-crypto-compliance.
• CoinDesk, on pump.fun acquiring Kolscan (11 July 2025): coindesk.com/markets/2025/07/11/pumpfun-acquires-wallet-tracker-kolscan-to-expand-onchain-trading-tools. DL News on trading before the announcement: dlnews.com/articles/defi/memecoin-kols-bag-millions-as-pump-fun-buys-kolscan-tracker.
• Kolscan daily leaderboard, read 21 September 2026: kolscan.io/leaderboard.
• The Block, on ZachXBT's paid-promotion documents (1 September 2025): theblock.co/post/368956.
• US Securities and Exchange Commission, press release 2022-183: sec.gov/newsroom/press-releases/2022-183.
• Cointelegraph, on Murad Mahmudov's talk and wallets (October 2024): cointelegraph.com/news/murad-mahmudov-memecoin-supercycle-spx-holdings-controversy.
• Orangie, a first-person account of trading history: x.com/orangie/status/1827727018286358731.
• SlowMist (July 2025): slowmist.medium.com/a-popular-solana-tool-on-github-conceals-a-crypto-stealing-trap-bfef0aed63d7. StepSecurity (15 March 2026): stepsecurity.io/blog/malicious-polymarket-bot-hides-in-hijacked-dev-protocol-github-org-and-steals-wallet-keys. Socket (8 January 2025): socket.dev/blog/gmail-for-exfiltration-malicious-npm-packages-target-solana-private-keys-and-drain-victim-s.
• BingX spot rates are as published and as used in the annual fee math post. Our own figures come from the Memecoin Desk snapshot of 21 September 2026.

Earlier posts that bear on this one: How we decide which tokens to cover on the checklist behind the seven coins above, and Do crypto signals actually work? on why a published track record means nothing until someone grades it.

Disclosure: RebateMax earns affiliate commission on trading fees generated by accounts registered through its BingX link and returns 50 of those 60 commission points as a lifetime discount. The full arithmetic is here. RebateMax also earns a referral commission from LBank. It has no commercial relationship with pump.fun, Axiom, GMGN, Photon, BullX, Kolscan or any person named in this post. Memecoins are extremely high risk and most go to zero. This post is general information, not financial or tax advice, and nothing here is a recommendation to trade. Think a figure is wrong? See the corrections policy.