Why is my pump.fun sniper bot losing money? Six leaks to check

Most pump.fun bots lose money, and most of their owners only find out from a shrinking balance. A balance does not tell you where the SOL went. This guide lists the six leaks we see most often in on-chain data, how to check each one yourself, and how to get the numbers for your own bot in one step with the free Test my bot page.

First, measure the result properly

Before tuning anything, you need one honest number: SOL received, plus the value of tokens your bot still holds, minus SOL paid, fees included. Two traps make bots look better than they are:

• Counting only closed trades. A bot that never sells a dead token never records the loss. On 17 October 2026, counting only fully sold positions made the pump.fun market look profitable in our data (+13% on the SOL staked); once unsold tokens were valued at the lower of their cost and their market price, the median active bot was down about 22%.
• Counting sales of tokens bought earlier. If your bot sells a bag it bought before the period you look at, the sale looks like free profit.

Test my bot applies both corrections: unsold tokens are valued at the lower of their cost and their market price (median of the token's last 20 trades of at least 0.01 SOL), so paper gains never count, and only sales of tokens bought inside the measured window are counted.

Leak 1: fees eat a thin edge

Every bonding-curve trade on pump.fun pays a protocol fee and a creator fee, on the way in and on the way out. In our data for 15 October 2026, the median protocol fee was 0.95% of the SOL amount and the median creator fee 0.30%. A round trip therefore costs about 2.5% before priority fees and tips. A bot whose median winning trade is +3% is barely breaking even.

Check: divide total fees paid by total SOL staked. If it is close to your average gain per trade, trade less often or only on larger expected moves.

Leak 2: unsold bags

Bots that buy many launches and exit only on a take-profit leave a tail of tokens that never recover. Those bags are real losses even if no sell ever happened.

Check: count open positions older than 24 hours and value them at the token's last price (or at cost, whichever is lower). Add a time-based exit if that value keeps growing.

Leak 3: entering too late, or exactly too early

Entry speed decides who is on the other side of your trade. Buying in the token's launch block requires advance notice or automation; buying a few seconds later often means buying from those who did. The hourly pump.fun weather shows, for the last 24 hours, the return of positions grouped by entry speed: launch block, 1 to 10 seconds, 10 seconds to 5 minutes, later.

Check: your bot's median delay between token creation and its first buy, and the share of its entries in the launch block.

Leak 4: buying into prepared launches

Some creators launch dozens of tokens a week, with the same wallets buying in the launch block each time, and sell everything within minutes. The weather page lists these serial launch teams for the last 7 days with how often the creator dumped within 5 minutes.

Check: filter out tokens whose creator appears in that list, or whose launch block already holds three or more buyers.

Leak 5: holding too long, or not long enough

Holding time interacts with every other leak: short holds multiply fees, long holds multiply dead bags. Compare your median holding time with your winning and losing positions separately.

Leak 6: a small sample that looks like skill

Ten winning trades in a row says little. A rank only means something once there are enough positions: Test my bot ranks a wallet only from 10 matured positions, against every other active bot measured the same way.

Get all six numbers for your bot

Paste the public address your bot trades from on Test my bot. The free result gives the verdict: net result in SOL, positions, share of winning positions, rank and last 7 days. Pro (0.1 SOL for 30 days) adds the why: fees paid, unsold bags, entry speed, share of launch-block entries and holding time. Never paste a private key or a recovery phrase: the site never asks for one.

FAQ

Is it normal for a pump.fun bot to lose money?

Yes. Among the 118,493 bots with at least 10 matured positions measured on 17 October 2026, the median one had lost about 22% of what it staked.

Can I test a bot that I do not own?

Yes. Any public Solana address can be tested; the data is public on-chain trading activity.

Does a good result mean the bot will keep winning?

No. The test describes the past. Market conditions change hour by hour, which is why the weather page exists.