I’ve been working on Replit for a long time now, but feeling frustration that its taking 25 minutes to do simple tasks in power and max modes. Is anyone else experiencing this?
When simple tasks start taking 25 minutes in Power and Max, it’s usually not the model being slow. It’s the agent re-reading a project that has grown. Every task pulls in context first, and past a certain project size that read costs more than the actual work does.
Two things worth checking. First, how big the project has got, including files the agent scans that it doesn’t need for the task at hand. Second, whether you’re continuing one long-running chat. Starting a fresh session for an unrelated task is often dramatically faster than continuing a thread carrying months of history, because that history gets replayed every time.
Did the slowdown creep in gradually as the project grew, or did it change suddenly around a particular date? Gradual points at your project size, sudden points at something on their side, and those have completely different fixes.
Ok this is incorrect, in the last several days processing time on the same application doing similiar tasks as tripled or gone to 5x in time. The agent (or subagent, unsure what they are now doing) is taking forever and has lost its context of discussion making generally simple mistakes, and not delving into the history, but jumping to conclusions and taking incorrect facts. I locked it to max OPUS 5 to stop this poor behavior and spent hours getting it back in line. Replit has significantly changed its operations. This is not a app issue, its a systemic change in AI management and its costing 4x as much.