Agent charging exorbitant amount to answer simple questions in plan mode ever since the latest agent change

I’ve noticed lately that the agent has started charging me crazy amounts of credits for answering simple questions in plan mode. I tried contacting support hoping to bring this to their attention that something has gone wrong with the new agent modes and hoping for a refund on the problem, but got a basic response of “the agent charges for the work it does” and don’t seem to be acknowledging how off this is compared to how it typically is for a plan question.

These are far different costs than I used to get charged in plan, typically this would be .02 or so. I’ve tried using the agent in different modes. All of these questions where in economy mode and the last one was even in lite mode.

Edit: To expand on this, i logged into another account that isn’t paid. I asked the agent the exact same question. It took 34 seconds to respond and only charged .05. Something is obviously messed up with the agent in my paid account. It’s pretty frustrating to be blown off by support about it.

1 Like

Im am having the same experience, asking a simple question cost me 1$ for “general work” this is messed up!

I’m not an official Replit staff member or support engineer.

I’m at work myself scrolling on a break…

You should try the following:

If you know what GitHub is

1.) make a ticket edit: make ticket with Replit sorry on my phone

2.) open the git option in the Replit console

3.) push to git

4.) plan to use vs code to pull from git or clone repo/ use co pilot

4.) add a few redundant safe copies / branches

5.) ask these small high level questions to the free agent on vs code or GitHub

Wait for the changes and bugs to work out :person_lifting_weights: bound to be kinks…

My best guess is that because we are basically renting a VM and using the Replit agent that way is we are not seeing the full picture.

Also, it’s though to know when one agent history may have a lot compared to your free plan. It might be doing more than surface level work.

Try starting in a new chat box on the paid version and if you still think it’s a platform bug just pause. Take it offline.

Manually debug (even using AI tools)

I came here to say the same thing. Plan mode charges me, on average, 10x the amount just asking simple questions than Build mode. I am no longer using plan mode and saving a lot of money.

I’ll bite…

Plan mode likely uses the most powerful model available to ensure a holistic code purview.

Build, on the other hand, can be hamstrung by the user to use low-power models that are still, frankly, perfect for answering basic questions.

It’s a question of token usage, nothing more, nothing less.