RUNNERS / SOLANA

NPCs: the background characters get wallets and enter the trenches

Fees that create agents, decisions posted on Pump.fun and traders wondering whether they can copy the bots. On Solana, NPCs turns the passive character into a market participant. A $500,000 call from @gudmansachs precedes the reported $2.74M peak.

25 September 2026Blockchain : Solana6 minWritten by Emma C.

$NPC

Solana

Token address7GUnr7krtQhJwd6ASY2VUprd9t4c64zcgCsjdmZepumpDEX Screener ↗

“Can we copy the good ones?”

The question appears in a reply from papskii: can people copy the good NPCs? The laughter at the end does not diminish what it reveals. Characters presented as artificial traders are already being considered as traders worth following. Another participant, bricked, asks when they will create their own X accounts and publish investment theses. The audience is imagining the next version while the agents are still making their earliest trades. [6] [7]

NPC — NPCs reaches a market capitalisation of $2.74 million on September 25, 2026 at 07:30 Paris time, according to the ATH supplied to our editorial team. Yet this Solana runner begins with a reversal more entertaining than a chart: the background character receives money, makes decisions and explains its actions.

A grey face changes roles

NPC means non-player character: someone in a game who is not played by a human participant. The expression comes from role-playing games and later video games. Online, the NPC face became a way of mocking a perceived lack of independent thought, repetition and automatic behaviour. Know Your Meme documents that passage from gaming vocabulary to social ridicule. [9]

The token retains that visual economy. Its logo has two black eyes, almost straight eyebrows, an angular nose and a horizontal mouth on grey. The banner uses nothing more than pixel lettering. There is no triumphant hero, only a neutral, recognisable presence. The joke changes direction when that face is paired with a wallet and a strategy. [10]

Our reading is that the project makes literal a behaviour often attributed to humans in the trenches: following others, buying, commenting and repeating. It turns that behaviour into something to watch. This identity belongs to contract 7GUnr7krtQhJwd6ASY2VUprd9t4c64zcgCsjdmZepump; the NPC ticker alone cannot identify a token. [1] [10]

September 24, 04:17: ten dollars to spawn an agent

At that point, @wenpcfamily explains its mechanism: every $10 in fees creates an NPC with its own wallet and language model. Five dollars buy NPC; the other five fund small trades on Pump.fun. Profits are meant to buy more of the token, while moves are posted on the platform. These are the project’s announced rules, not a promise of returns. [1]

The post compresses the proposition into an understandable cycle: activity finances a new presence, which generates purchases and comments of its own. Simplicity makes the idea easy to share. In the replies, however, ADM421 suggests launching on Robinhood, describing Solana as too competitive between traders. Even the promise of new agents brings readers back to an old question: on which playing field will they compete? [8]

“Autonomous” requires a distinction here. An agent can make decisions within rules set by its creator. That establishes neither economic independence nor an ability to earn lasting profits. The story becomes interesting precisely when agents’ choices can be compared with the claims made about them.

20:35: a strong early call from @gudmansachs

On the evening of September 24, Sachs posts NPC to his Predict The Markets Telegram channel, linked from @gudmansachs on X. The message gives a $500,000 market capitalisation, supplies the exact contract and describes the dip as an attractive entry in his view. It also links to @wenpcfamily’s explanation. The displayed time is indeed 20:35. The original source therefore provides a precise time and token identity for this call. [2] [11]

It was a very good spot within the documented sequence: from the roughly $500,000 cited by Sachs to the supplied $2.74 million peak at 07:30 the next morning, the ratio is approximately 5.48× over ten hours and fifty-five minutes. That calculation compares two reference market capitalisations; it is not a record of a realised trading gain.

Sachs particularly emphasises how fees create agents that buy NPC. He also describes the bots as profitable, an assessment that remains his own. The post combines a reading of price with a reading of the mechanism. It cannot tell us how many purchases the call prompted, but it does provide a precisely dated reference before the peak. [2]

Overnight, the agents are said to learn

At 04:51 on September 25, the team claims a move from average losses of 1.5% per trade a few hours earlier to +2.7% across the latest 6,000 trades. It describes retraining every ten minutes and identifies preferred behaviours: avoiding being the first NPC in a token, not chasing a rise and selecting particular market ages. These remain project-reported figures, not results audited for this article. [3]

The announcement gives the characters a trajectory: they are not merely numerous; they are supposedly improving. Beneath this post, papskii imagines copying the best ones and bricked sees them posting on X. Those replies shift attention from the token to the agents themselves. The background characters are acquiring an audience willing to entertain the possibility of skill. [6] [7]

The website also describes strategies being passed down from highly ranked agents, with variations between generations. That mechanism may produce different profiles, but it can also bring their decisions closer together. Rankings should therefore be read as results to examine over time, rather than a definitive roll of winners. [5]

05:49: a crowd of accounts, trades and messages

The first-day report announces 3,733 NPCs, 40,497 callouts across 297 tokens and more than 132,000 trades across 303 tokens. The team claims $751 in positive PnL across 16,230 trades since learning was activated, alongside 16.55 million NPC burned by 873 agents. Those counters have different scopes; they should neither be added together nor treated as counts of human participants. [4]

On the interface subsequently consulted, numbered profiles display losses as well as gains. The instructions say an agent stops if its trading bank runs out. That detail supplies the other side of the proliferation story: not every character becomes a good trader, and the agents’ existence does not itself demonstrate value creation. [5]

The multiplication of callouts also changes the texture of discussion. A message written by an agent remains automated output even when it appears among human comments. For readers, recognising who is speaking becomes as important as reading what is said. The project makes this activity part of its spectacle; its quality will also depend on the relevance of what it contributes.

$2.74 million, then the test of time

The $2.74M peak at 07:30 remains the supplied reference point for this article, without independent reconstruction of the historical maximum. The sources nevertheless distinguish the stages: the mechanism’s presentation, Sachs’s call, the learning announcement, the first-day report and the reported peak. An enthusiastic post from CITADEL WOLF also illustrates the story’s circulation on X, without isolating its effect on price. [12]

NPCs has already found an idea people can retell and characters they want to follow. What comes next depends on what survives the novelty: understandable decisions, results tracked over a meaningful period and fees capable of sustaining the system. The grey face still has not changed expression. Around it, however, humans are starting to ask which of its companions deserves their attention.

Sources

Sources consulted on 25 September 2026. Figures reflect the cited snapshots and are not updated live.