Short answer: A GMX perp position pays position fees on open and close, a borrowing fee that depends on pool utilization, a funding fee that depends on the long-short imbalance, and a price impact that is applied when you reduce or close. A bot needs all four in its cost model, because the holding costs scale with time.
Costs that happen once versus costs that accrue
Position fees are charged when you open, close, increase or partially decrease a position: 0.04% or 0.06% of position size, with the lower rate when the trade moves the market toward balance. These are the same on every trade and easy to model.
Borrowing and funding accrue over time. A strategy that holds for minutes barely notices them, while one that holds for days can have its edge eroded by them.
Borrowing versus funding
The borrowing fee is paid by open positions and falls on the side with more open interest, and its rate changes with the pool utilization percentage. Funding is a transfer between longs and shorts: when longs outnumber shorts, longs pay shorts, and when shorts outnumber longs the direction reverses.
That means funding can be income or cost depending on which side you are on. Positive funding can be collected from the claimable funding box in the interface.
Price impact and the network fee
Trades that improve the long-short balance get a positive price impact and trades that worsen it get a negative one. According to the documentation, nothing is applied when opening, and the net impact is applied when the position is reduced or closed. If the net impact on a closed trade exceeds a 0.5% threshold, the excess can be claimed as a rebate after 5 days.
Orders are executed by keepers, and the interface charges a maximum network fee that is deliberately overestimated for gas changes. Any excess is returned to your account.
Turning it into a cost model
For each strategy, estimate the expected holding time and open interest conditions, then sum position fees on entry and exit with the accrued borrowing and funding for that holding time. If the expected gain per trade does not clearly exceed that sum, no execution trick will rescue it.