In June 2026, CCN published a line that should have made every “creator economy” analyst sit up: pump.fun — a website where you launch a meme coin in under a minute — had paid its users over $350 million in creator fees, most of it during a bear market. Not to venture funds. Not to the platform. To random people who clicked “create coin,” picked a name, and got a cut of every trade on that coin forever.

At the peak of the fee model in September 2025, creators were collectively earning about $2.1 million per 24 hours — a run rate that had individual streamers out-earning Twitch partners, on a platform whose own history includes a livestream feature it had to suspend over people running Russian roulette on camera.

Then the abuse started. And it got bad enough that by January 2026 pump.fun publicly admitted the creator fee model had failed, and spent the next ten months rebuilding it — capped redirects, cashback coins, holder rewards, the works.

This is the full post-mortem: how the payout machine actually worked, the exact numbers, how it was farmed, and what the rebuilt version looks like in October 2026. Everything below is sourced from on-chain reporting and primary coverage — and where a number is illustrative math instead of a verified figure, it says so.

The machine: how creator fees actually worked

Pump.fun launched on January 19, 2024 as a launchpad for Solana meme coins. Anyone can create a token for less than $2 in total fees. The token starts on a “bonding curve” — an automated pricing curve inside pump.fun itself — and if the market cap hits roughly $90,000, the token “graduates” to a real exchange pool on PumpSwap, pump.fun’s own decentralized exchange. Since the platform’s fee share was widely reported as roughly 1% of trades, that graduation event alone was worth 1.5 SOL per coin to pump.fun (~$154 at the October 24, 2026 SOL price of $102.44).

The creator fee is the part that went to you, the token’s deployer:

  • On the bonding curve: the creator earned 0.30% of all trading volume on their coin.
  • After graduation: the creator fee became variable, ranging from 0.05% up to 0.95% of trading volume depending on which market-cap tier the coin occupied.

Read that again, because it’s the whole thesis: not a one-time payout, not ad revenue — a permanent royalty on trading volume, paid in SOL, for a token you created in one minute. Volume, not price direction. A coin that churns $10M of trades hands its creator roughly $30,000 on the curve alone, even if the chart is a disaster.

Scale context, all verified: by January 2025 the platform had hosted over 6 million tokens (per Wired), and pump.fun has generated nearly $800 million in cumulative revenue for itself. In July 2025 its token sale raised $1.3 billion in about 12 minutes of public-sale time. The creator fee program was, until the abuse arrived, arguably the largest experiment in paying anonymous internet users for content — except the “content” was a token and the payment was an on-chain royalty stream.

The peak: $2.1 million a day, more than Twitch

In early September 2025, Yahoo Finance and Benzinga reported that the fee structure was generating over $2.1 million in creator earnings per 24 hours, with prominent streamers on the platform out-earning top Twitch creators. For perspective: a Twitch streamer with 5,000 concurrent viewers typically makes a few thousand dollars a month from subscriptions. A pump.fun creator with one genuinely active coin was collecting a slice of every trade, 24/7, with no platform holding the funds, no payout threshold, no “your account was flagged” email.

It’s worth being honest about why the money was there, because this is where most takes stop: creator fees are paid out of trading fees, and trading fees are paid by traders. pump.fun’s own data, reported by crypto.news in March 2026, showed 49% of March traders in the red. The $350M creator pool is real money, but its source is a much larger pool of aggregate trader losses. If you build a business on fee-sharing from a casino, you should understand you’re not building an annuity — you’re participating in a churn economy, and churn economies don’t do stable payouts.

How it got farmed: three exploits of the incentive

Any system that pays a recurring stream to an anonymous wallet based on activity will be farmed. Creator fees had three systemic failure modes, and all three eventually made headlines.

1. Volume farming through self-trading. Since the fee is a percentage of volume, the cheapest way to “earn” it is to trade your own coin back and forth. Every round-trip costs the trader’s fee — but a portion comes back to you as creator. Wash-trading your own bonding curve to farm your own fee stream was the obvious move, and at scale it turned the creator program into a subsidy for fake volume.

2. Fee-stream hijacking — the Kimchi case. This is the case that broke the model’s reputation, and it’s fully documented. On October 20, 2026, a token called $KIMCHI launched on pump.fun. The deployer’s very first transaction bought about 79.3% of the one-billion supply for roughly 87 SOL (~$9,000–$9,300). Minutes later, 750 million tokens (75% of supply) landed in a wallet labeled “Kimchi” — the handle of a trader who had just been announced as a pump.fun collaborator on creative direction — and the deployer hard-coded that address as the permanent, unchangeable recipient of every creator fee the coin would ever generate. Promotion tied to the name pushed the market cap to about $40 million within an hour. The deployer sold the remainder for an estimated $26,000 profit; snipers did better. The fee stream — a revenue stream on a $40M-cap coin — was permanently routed to a wallet whose control nobody could verify, minutes after a coordinated supply grab.

Pump.fun’s answer, reported by Cointelegraph on March 25, 2026 (after earlier incidents and after the January pivot): deployers get one post-launch change to fee settings per token, then it’s locked forever. One redirect, not unlimited ones.

3. Attention hijacking. A September 2026 incident reported via TradingView/U.Today: a platform called UsePaid — which lets creators route payouts to X (Twitter) money — paused payouts after a pump.fun launch surge sent creator-fee claims to $1.54 million in a single day. New tokens were being deployed at a rate that turned third-party payout rails into a firehose, and the deployers behind them were farming exactly the volume-fee loop described above.

The pattern in all three: the creator fee didn’t pay for content or community. It paid for attention and volume, and both can be manufactured.

The rebuild: January to October 2026

To its credit, pump.fun didn’t just let the program rot. The rebuild happened in three visible phases:

  • January 2026 — admission and overhaul. Coverage by The Block, Cryptonews and Yahoo Finance on January 9–10, 2026 describes pump.fun overhauling creator fees because, in the platform’s own framing, the old model enabled “dangerous” low-risk farming. Token launches hit their highest daily count since September the same week — the redesign shipped at peak activity.
  • March 2026 — the one-redirect cap. The fee-recipient lock described above: one change per token, then permanent. This specifically kills the fee-redirect rug, because you can no longer flip the recipient around post-launch; whatever you set is what you’re stuck with.
  • October 2026 — the pivot away from creators entirely. The most recent phase, and the reason this post-mortem matters now: pump.fun redirected fee value to traders. CoinMarketCap reported the “Cashback Model” on October 21, 2026; Binance’s item is blunt — it targets “token deployers who don’t deserve the fees.” Later in the month, pump.fun went further: cashback removed as an option for new launches, replaced with Holder Rewards — automatic payouts to wallets that hold, instead of a trading-linked rebate. Early effect, per The Block’s reporting: token graduations hit a seven-month high, with weekly graduation rates around 1.15% of all launched tokens, the highest since the previous summer.

So the arc in one paragraph: pump.fun invented a permanent on-chain royalty for token creators, paid out $350M+, watched the incentive get farmed by supply-snipers and fee-redirect rugs, capped the exploitability in March, and by October concluded that even the capped version rewarded the wrong people — and shifted the economics to the people actually holding the bag: the traders and holders.

What still pays in October 2026 — and what doesn’t

If you’re looking at this as a monetization case study, the honest ledger:

Mostly dead: launching random coins to collect creator fees. The October pivot explicitly moved fee value from deployers to traders/holders, the redirect cap killed fee-stream games, and the Kimchi coverage put deployer-side abuse under permanent scrutiny. The 2024-era playbook — mass-deploy, farm curve volume, repeat — is farming a shrinking, monitored pool.

Still real, but different: genuinely engaged token communities. The graduated fee tiers (0.05%–0.95%) still exist, and a coin with a real, active holder base still generates a creator stream under the new holder-reward structure — the payout just now also flows to holders, which is both a cost to the creator and an incentive alignment that the original model lacked.

The durable lesson — and the actual takeaway. The $350M experiment proved something important: on-chain royalty streams are programmable, permissionless, and instant. No payout thresholds, no KYC’d platform, no 90-day net terms. Whatever you build — content, tools, data, tokens — the mechanism of “a permanent programmable cut of activity, paid to a wallet, in real time” is now table stakes. pump.fun’s failure wasn’t the mechanism; it was attaching the royalty to the easiest thing to fake. Attach it to something hard to fake — actual usage of software you wrote, actual audiences you own — and the same rails that paid $350M work in your favor. The people who lost out in every phase of this story were the ones who confused the mechanism with the moat.

The sharp edges

The educational disclaimer at the top of this article is doing real work here, but some specifics deserve their own paragraph:

  • Most participants lose money. Nearly half of pump.fun traders were in the red in a single tracked month, and creator payouts are funded by aggregate trading losses. This is a negative-sum environment for participants as a whole.
  • Regulatory heat is real. pump.fun banned all UK users in December 2024 after an FCA warning, and a US securities lawsuit (filed January 2025 in the Southern District of New York) claims the platform operated as an unregistered securities exchange. Earning “creator fees” from token launches is exactly the kind of activity that regulators read as issuing unregistered securities — jurisdiction determines your exposure, and you don’t get to pick after the fact.
  • Snipers see you coming. The Kimchi deployer’s 79% supply grab was replicated by an ecosystem of bots that snipe bonding curves within seconds of launch. Any strategy you read publicly is priced in by software faster than yours.
  • Payout rails break. UsePaid paused $1.54M in claims because volume spiked past what its systems could handle. Third-party plumbing in this ecosystem is fragile.
  • Nothing about this is passive. “Passive income” coins are a marketing category, not an economic one. Every phase of pump.fun’s fee economy rewarded active, information-advantaged, often automated participants.

Bottom line

The creator fee program was simultaneously the biggest anonymous-creator payout in crypto history and a controlled demolition of its own incentive design. $350M+, $2.1M a day at peak, three systemic exploits, one hard-coded fee rug that made every headline, and a ten-month rebuild that ended with the platform paying traders instead of deployers.

The numbers are real. The mechanism is real. And the post-mortem verdict is the most useful part: programmable on-chain royalties work — right up until the thing they’re attached to can be manufactured. Whatever you build next on these rails, attach the fee stream to the part of your project that can’t be sniped in the first sixty seconds.