I have issues with Git as well, I read it could be due to the size of your project, mine s over 2gb in code and I don’t know how to push it now that its such a large project
1 Like
Thanks @jefftj86. I suspect it could be the same thing. I hope to god that Replit can fix it and quick!
Help is on the way. Hang in there
1 Like
@Gipity-Steve Thank you for your effort in aggregating this into a singular post; you’re a saint.
I was astounded by the drastic differences in costs for what was costing much less only recently. After the first price hike, I raised similar concerns to Replit, especially about a noticeable difference in quality and results. I received Bot responses and eventually some amount of compensation, whose calculation was ambiguous.
Pricing Model is Logically Broken:
- In any case, I’m only going to voice the following because clearly I (and I bet “we”) are passionate about what Replit has to offer, but we also recognize the pricing model is far from ideal or sustainable.
- Usually, I charge for Product/Pricing consultation, and what @Gipity-Steve is doing is essentially Product Management too, but I digress:
- The pricing model is correlated to cost more, incentivizing the platform to make more mistakes, give us dumber Agents, and take the Billy path to a solution.
- As you noted, this drastically erodes user trust in the platform and/or the company. “Agent 3” doesn’t seem any smarter, but much costlier.
- This begs the question: are the creators of Replit disguising pricing hikes as platform enhancements?
1 Like
I am new to the platform. Approaching my first month anniversary. For the first 2 weeks I seemed to work for hours and barely depleted my $40 credit. After the Agent 3 release, I blew past the $40 and more than doubled it in 2 days. This amount may be small to some of the enterprise users out there but the concern is the same. I thought I would be able to develop apps on a continuous basis around the $100 mo but that seems highly unlikely. If the trend continues, I will have to park this and look at other options.
I spent hours with a preview that would not show. (What good is coding if you cannot see the output). Support said it was a know issue? Honestly - the support looked AI generated. Too many “Great Job! You found the issue” comments to be a real human. Overall, very disappointed in the experience. Standing by for the outcome but exploring other options.
2 Likes
To be clear, you could never develop apps continuously for $100/mo. - not even in the early days. But it was much much cheaper not long ago.
2 Likes
Major apps for sure. My use case was simple landing pages, Jamstack sites and form processing. It was doing a great job a few weeks ago when I started this journey, but then what used to cost $2-$3 or so on a project turned into $30. I am just saying, even though my usage may be small, I can see the proportion increase has been exceptionally high. Instead of my original $100 month usage estimate, I am now looking at $300-$500 which was a surprising change!
Ray,
Your use case is too common I’m noticing.
And yes, the responses you get from the platform are a bot. A well trained bot of course. How else can they provide you a response within minutes?
I think at the end of the day the platform wants to make it profit and scale its growth. But it needs to figure out how to find the balance.
This means less experimenting on our part as users because it’s gonna cost too much. Now the question is is that lower usage going to be profitable for Replit?
In other words, is it more costly to allow users to experiment for a low cost or is it more profitable to throttle users use by putting a high price tag on things? It’s very simple economics.
lost the agent running chat cost. that’s rough. i want to know where i’m at per project, per chat.
kind of gives me bad vibes.
3 Likes
Yes this is critical for me too. @FranciscoCM - can we please get back the little hover-over icon that shows the running total cost for the current chat.
1 Like
Plan / Build
I love plan mode and build mode - helps stop major un-wanted drifts in code. But here’s where costs seem to really inflate w/o need.
I send a prompt to plan. It plans. I say “do it”. Then it takes 22 seconds to say “ok, put me in build mode”. But at a cost of .12 cents.
Look - put some basic filtering in before we send off to the agent maybe? “Go” “Proceed” “Build” - a bunch of things that sound like ‘yes’ could trigger a “do you need build mode”? Yes/No - and avoid all the compute with LLM.
I know it’s just 12 cents. But that’s $21.60 an hour to tell me to turn on build mode.
Of course - i can learn my lesson…but hey, we’re all about building software for users here, right?!
2 Likes
This 1000%
Further -i know it’s been discussed. But i’m actively using Replit and running into this…see my comment above.
22 seconds to tell me to turn build mode on: 12 cents.
Then i turned it on and said “Go”
And it went. Yay!
But 2 minutes of work : 60 cents
I understand this isn’t straight pricing per second…but then what is it? I think i’m reading the assistant is only 5 cents per file edit.
I paid 5x this time for a 80 lines of work.
But above paid 12 cents for chat - that really had no value.
I’m curious if we could get metrics on our ‘chat’ costs vs ‘work’ costs (and is that build vs plan). And from there, i’d really be curious if the two models are priced differently somehow.
And seriously - now that we’re all noticing the price increases post Agent 3, could a lot of this be the planning? OR - is this the baked in ‘extended thinking’ that used to be an option?
If so - is that what my agent now refers to as 'the architect’ ? has anyone else seen this - i can’t find reference to this in the docs. Sorry if i’m slow.
I’m all about improved performance - but would like to understand how it’s happening, and where the costs are coming from.
3 Likes
Switching all tools on and off - including Agent 2/3 modes
Something @realfunnyeric said makes sense:
I agree that the most logical and expedient solution would be to allow the user to select either Agent 2 or Agent 3.
But not just as an emergency fix. We need it long term.
So just like we have the different extended thinking/power modes, I believe all these new sub-agents and tools should be things we decide if we want enabled.
Switching which LLMs are used
Including control over which LLM models are being used - I suspect Architect is the bank-account killer, as it seems it is automatically hardwired to use the top Claude model - which costs more than my house to run.
Complete re-think & re-design of the tool/model switching UX
If Replit do go down this route of giving users better choices, then make the UX much slicker:
For example, the way we currently move between Plan and Build modes is very clunky - when agent is instructed to build but finds itself still in Plan mode, it charges $0.12 just to ask us to manually switch.
The whole Agent/workspace user experience needs a lot more design.
2 Likes
Account management for Core and Teams is a mess
This isn’t so much an issue related to Agent 3, but:
Am I the only person who finds the whole account management confusing and poorly thought out?
Core vs Teams; personal accounts; organisations; teams; groups; members;…
I am totally lost!!
I had an old Core account but no longer pay for it so can’t use it. But it seems I still login with that one simply to get to my new Teams account, where I am a member in group. The Teams account is the one with my credit card and that pays, and I think has a different user profile, but my invoices go to the email associated with the old Core account. So confusing!!
And as for the Teams roles/permissions - they are ridiculously over-complicated.
Also, unlike the Core account, I can no longer invite other users outside my team to collaborate, or be invited to collaborate on their Core apps. These are critical requirements for those of us who support other Core (or external Teams) users as their developers or vibe coding wingmen. (And seperately: if we can collaborate on an external Core user’s app, it would be good if we could publish their apps.)
Also, it seems chat between collaborators has been removed since agent 3.
And let us not forget the legal issue related to Core accounts: that we grant Replit some control/ownership over our Core apps; so those who commercialise their use of Replit MUST switch to Teams,… and thus lose some of the simpler features that allow collaboration with the very external users who they are commercialising with!
So, overall, an utter mess.
The only solution for my company to stay with replit is if
- they offer Agent 2 as an optional setting. it worked for OpenAI as a stop gap while they fixed gpt5, it stopped serious declines in users.
We also need:
- urgent simplification of the pricing. and
- free egress fees (through replit using cloudflare integration for file storage and downloads).
1 Like
@Marshal_Thompson made a comment elsewhere which reminded me of my concern with pricing and this whole “agent 3 can run for 200 minutes”:
Essentially, there is no way on God’s green earth I am letting an agent run for hours at a time. 10 mins is about all I feel comfortable with.
Someone else on here used this analogy, and it’s a good one:
Like getting in a taxi in a new city, and the driver refuses to tell you how far away your destination is, how long it will take, or what the billing model is (is it per mile, or measured by time, or both?). AND, you have to give him your credit card at the start of the journey and sign an agreement that says he can bill you whatever he likes once you arrive. Oh, AND, you have no guarantee he will actually take you to the right destination. If you arrive and it isn’t the right place, you agree to pay him more to try again.
This model is utter nonsense, and if Replit continues with this concept, I will be very worried.
2 Likes
Documented Issues with Current Agent Behavior
1. Memory Loss & Context Switching
-
Agent loses context within seconds of new task assignment.
-
Previously completed work is not retained between interactions.
-
File discovery takes 10–15 minutes versus ~10 seconds with Assistant.
2. Task Execution Failures
-
Roughly 80% of tasks are affected across all model types (orchestrator, high-capacity, basic).
-
Agent frequently stops mid-task without completion.
-
Code audit requests consistently fail.
-
Architecture creation requests often loop indefinitely, consuming compute credits.
3. Performance Degradation Timeline
-
Problems began after introduction of the orchestrator model.
-
Overall performance has degraded by ~90% compared to earlier versions.
-
Issues persist even after using “kill 1” or starting fresh chats.
4. Billing Impact
-
Each bug incident consumes $200–300 in compute.
-
Past incident included a $51 charge for a stuck agent session.
-
Estimated 60% of total billing directly caused by agent failures.
Technical Environment
-
Application with 4000+ files and multiple service integrations
-
Existing performance monitoring and error tracking systems in place.
-
Heavy real-world usage: 16 hours daily, $4k+ spent.
Code Quality Issues
-
Agent-generated code contains recurring bugs and inconsistencies.
-
Repeated failure to properly read and analyze the existing codebase during audits.
-
Integration attempts produce broken or misaligned architecture.
Recommended Immediate Actions
-
Escalate systemic issues to the Agent v3 engineering team for urgent investigation.
-
Implement billing protection for looping/stuck sessions to prevent runaway charges.
-
Provide temporary compute credits for users impacted by repeat failures.
-
Consider rollback to a previous stable Agent version until reliability improves.
1 Like
I am having these issues as well. I have a very large system as well, Thousands files and gigabytes of code, agent 2 manages working with this with no issues. I had a good flow we knew how to work around context limits and overall figured out how much everything would cost, I averaged $1000-$1100 per month but I use it every day hours of non stop development per day, around 3-5 hours a day. Now it doesn’t do anything, it has been stuck for over a week. This is crazy.
im also on pretty same stage , i have burnt 3k$ a month and they have messed up and when i mailed them and caught them with their issues , they told me we have no refund policy , some one need to sue them exploiting money of end user by running random test on their beta models , they are just capable of maybe just making snake games or calculators