It feels amazing to see the Agent 3 think through and act in depth, but it is painfully expensive. For small 1-2 mins thinking tasks as well it eats up a couple of dollars.
PFA screenshot of a large implementation done by Agent 3 yesterday (worth $32.60) - this is without using any high power model btw.
Rollback would be very painful 
same here
Okay, then you need to be quite sure about the outcome.
And I feel a bit reluctant to try it out, if it costs $30, just to try it once.
Based on how many rollbacks I have done with the Agent and Assistant.
It wouldnāt cost $30 for smaller tasks. You can surely try it out.
It costs $2-3 for small features. At the moment, it costs about $.50-$1.00 for almost any prompt through Agent.
Itās very difficult to know what a prompt will cost until its done.
Thank you for your feedback and the screenshot. Iāll be voicing your concern.
Same problem. I had planned changes with plan mode and the opus model, and asked it to do the tasks in a sequence. It ran though all of them in 3:30 hours and 40 dollars with a single rollback point in the end. All or nothing. And the changes did not result in successful features at the end, the app is now a mess and it just burned through 40 bucks.
Incredibly expensive. Basic fixes and feature implementations that used to take <5 minutes and cost $1 or less now take 15-20 minutes and cost $5 or more. In several cases, it has not fixed the issue or implemented the feature how Iāve wanted and has drained a lot of money for nothing. I even asked a basic question in "planā mode about why it is charging me so much for basic questions and it still charged me .32 cents for even asking that question! I unfortunately may have to bail unless some changes are made. The $25 monthly allocation is basically gone within a few hours if you are working on more complex things/arenāt incredibly careful with what you want it to do.
Yeah, I didnāt really think of it until I saw this post. My initial response was āYou can roll back to any of those checkpoints.ā The difficulty there is the development happens in sequence, so if the feature/fix you were excited for happens to be on checkpoint 1 out of 9ā¦ thatās a tough hit to take.
Agent & Assistant Same here, I spent 300+ USD for one day (yesterday) coding in 3 different apps! 
THIS is not acceptable 
Besides the already exorbitant cost of Agent 3 in Build mode, when I make edits in Edit mode the Agent repeats those same changes in Build mode and charges me again. This isnāt progress ā itās a reason to cancel my Replit accountā¦
please summarise this issue in Summary of Replit issues - ALL MEMBERS, PLEASE CONTRIBUTE ASAP
Same experienceā¦ Probably not much point repeating. At least I am glad I am not alone. I darenāt do a prompt - I would rather waste money on the lottery.
I spent a lot of time and money learning replit, given many demos to friends, but like many, am stuck given these changes. I feel like I am back at square one, want to get back on track, but the cost of experimenting to work out how to use it productively is now 5-10X what it used to be, so I cannot move forwards.
As a user, you literally have no idea if a prompt will cost $1,$10,$100, and whether it will work or not and so whether you have to go through the cycle over and over again.
replit used to advocate small incremental changes, as the way to use Replit. This was the workflow I learned to use and it worked well - I felt in control. Now I do not. Is this still even possible? If so, what are the approaches to use.
replit guides were awesome, but I am not sure if they are still relevant. e.g. is this still even correct in the context of Agent 3?: https://docs.replit.com/tutorials/how-to-vibe-code
Or is Replit now just for building complete small apps from a prompt, with escalating cost of making tweaks down the line?
If thatās the case, I am done, as you literally have no idea what an app is going to cost. Iād rather use a human who knows how to use cursor, and fixed price set of business requirements, which I can draw up using ChatGPT / claude.
Plus 2 5X price changes in 3 months, both delivered woefully badly. This does not bode well.
Hi, the same thing happened to me, almost $400 in 2 days, payments every 2 hours of work! I looked at my past performance and on average, I maxed out at 65,000 tokens in a day. Now, in two days, I easily surpassed 130,000 tokens, practically doubling costs with ease! Iām testing dyad and Iām getting the same feeling I had at the beginning with replit! Obviously thereās a bit of work to do, but my experience with replit over the past few months is helping me.
Same, Agent 3 spent $30 yesterday on a simple app to take a photo and share in social media, and the app doesnāt even work.
Same here - what makes it worse is being billed after finding the Agent didnāt even do what is said it did.
In addition, Iām finding Agent 2 was much better at identifying and fixing errors. Agent 3 doesnāt seem to accurately assess issues, then jumps to a half solution that doesnāt address the problem.
The testing is a nice idea, but itās not ready for prime time. Would love to rollback to Agent 2.
You are right and it seems to be a recurrent issue in replit as for every major change it starts to bill incredibly expensive and go back to normal billing and then when it is fixed a new cycle starts. To be honest tired of this billing issue. Maybe someone in replit can fix it !
I am new in commenting in the forum so be easy on me 
Wow, what an incredibly simple, but clear analysis.
So basically either Agent 3 is:
