Polymarket has fees now. Your copy-trading math just changed.
For years, "Polymarket has no trading fees" was a safe sentence. In 2026 it quietly became false — twice. The international exchange rolled out category-based taker fees (Fee Structure V2, effective March 30), then raised sports from 0.03 to 0.05 in July; and the separate CFTC-regulated Polymarket US exchange switched on fees exchange-wide from July 1. The authoritative schedules live at polymarket.com/fees and docs.polymarket.us/fees; the mechanics and the full category table are in our new fees guide.
This post is about the narrower question: what the change does to copy-trading math specifically.
Copiers pay the fee by construction
The international schedule is taker-only — makers are never charged. That asymmetry is the whole story for copying: a copy-bot watches a leader's fill and crosses the spread to chase it. It is a taker on entry and a taker on exit, every time. The leader might have earned the maker side of the same trade; you structurally can't.
The fee itself is C × feeRate × p × (1 − p) — it peaks at 50¢ and vanishes toward the extremes. On a sports market (rate 0.05), a share bought at 50¢ costs 1.25¢ in fee; closing it at 70¢ costs about another 1.05¢. Call a mid-price sports round trip roughly 2–2.5¢ per share against a 50¢ entry — before the bot's own fee, before slippage, before latency.
Stack it and compare it to the edge that actually exists
An execution bot commonly charges about 1% per side. Add the platform's taker legs and a realistic slippage-plus-latency cost and the hurdle a copied wallet must clear sits meaningfully above 2% per round trip on mid-price sports fills.
Now put that against the field. In our July 2026 snapshot of 1,649 actively-traded wallets, the median fee-adjusted edge was −2.7%, and 74% of wallets were negative after costs. The copyable tail — wallets clearing every test at once — was 22 wallets, 1.3%. Fees didn't create that distribution, but every basis point of fee moves the bar the survivors have to clear, and thins the tail further.
Three practical consequences:
- Headline return is now even more misleading. It was always computed on the leader's fills at the leader's prices; now it's also computed under a fee the leader may partly avoid (as a maker) and you can't. The 20–40% overstatement is the optimistic case.
- Price level matters. The fee is largest exactly where copy-bots do most of their business — contested mid-range prices. A wallet whose fills cluster near the extremes carries structurally lower fee drag than one that lives at 40–60¢.
- Category matters. Sports fills pay 0.05, crypto 0.07, and geopolitics currently 0. Two wallets with identical gross edge in different categories no longer net the same.
What we changed on our side: nothing, and that's the point
CopyGrade's realistic edge has modeled fees, slippage, and copy latency since the first Copy Score shipped — the model never assumed the fee-free world. The methodology prices a copier's round trip, not a leader's headline. So scores already reflect the regime above; what the 2026 fee schedule changed is how wrong the unadjusted numbers everyone else quotes have become.
If you take one action: stop reading gross PnL, and check what a wallet's edge looks like after the costs you'd actually pay. That's the number on every Copy Verdict — or run the simulator on a wallet's real fills with the fee knob set to what you'd really face.