I’ll weigh in on Agent 3 costs after spending more time with it, but I think the community’s main concern is valid: interactions like this should not cost $0.15. Most major LLM APIs charge only fractions of a cent per token.
Is Agent 3 defaulting to a high-cost model like Opus for basic chats?
Agent 3 should leverage a smaller model with a lightweight system prompt to triage casual Q&A versus complex tasks that should be handed off to the orchestration.
Paying $0.15 just to correct Agent 3 feels unreasonable.
I’m going to play devils advocate… why is it wrong? They can charge what they like - they are in business to make a profit. You are not using a direct LLM API, you are using a tool that sees and understands your code and app - there is a tonne of added-value in that.
The issues we’re all screaming about, IMHO, are more about lack of transparency:
Letting agent go off on wild goose chases. Assigning work to sub-agents (architect, testing, etc) without our consent. Agreeing to make one fix but then suddenly “fixing” 10 other things without our permission. Hiding the current chat’s running total cost so we don’t have any insight into our current spend. Not letting us plan our costs better - just saying its going to disappear for 200 minutes and surprise us at the end with a big bill for code that went down the wrong rabbit hole 5 minutes into its run.
But simple “I’ve had to do 11 seconds of work, consuming compute time, LLM tokens… so I want to charge you an upfront $0.15 for that” - I don’t know, that feels fairly open to me.
I’m only saying this, because I feel, at some point, we have to come to the table too, and accept that Replit somehow has to find a solution we can agree on. I fear many people (including those on Reddit\replit which I have just discovered today) will only be happy when Replit is free.
I hear you on playing devil’s advocate, but this isn’t about whether Replit can charge — of course they can. The point is whether the pricing model reflects the actual value delivered.
Charging $0.15 for what amounts to a lightweight correction or a single Q&A exchange is wildly disproportionate to the actual compute and token cost. We’re not talking about “11 seconds of work” — we’re talking about a trivial call that, under direct API pricing, is a fraction of a penny. Replit choosing to wrap that in a $0.15 minimum fee isn’t “open,” it’s just opaque mark-up hidden under the guise of convenience.
You mentioned the real problems: lack of transparency, runaway agent behaviors, hidden spend until after the fact. But those issues are made worse when even the simplest, low-value interactions are priced at a premium. Instead of triaging basic Q&A through a smaller model and escalating when orchestration is needed, we’re effectively paying top-shelf rates for tap water.
Nobody reasonable is asking for Replit to be free. We’re asking for proportional pricing, visibility, and a system that uses the right tool for the right task. Until then, $0.15 per trivial interaction isn’t about “profit” — it’s about misaligned incentives and poor cost engineering.
Hhmmm, I disagree with part of this (probably, but I may well be wrong, as this entire Replit upgrade is such a mess).
You are saying “Replit choosing to wrap that in a $0.15 minimum fee isn’t “open,” it’s just opaque mark-up hidden under the guise of convenience” and “misaligned incentives and poor cost engineering”.
Others would simply say it is “profit”.
You can agree or disagree with the levels of profit, but the market will ultimately decide, i.e. we will all move to Lovable - until they do the same. So then we’ll go to Bolt,…
Bottom line is, we’ve been sucked in to these new tools and now we all scream when they start turning the screws.
The same thing happened with Uber. Cheap taxis subsidised by VCs. And then when we were all hooked, the rates starting going up to “real” costs.
Or the drug dealers who hang out at the school gates, giving free drugs to first time users. And then when they are hooked,…
We fall for it every time. The AI Dev game is no different.
I get the analogy, but it misses the point. Nobody here is shocked that companies need to make a profit — that’s obvious. What we’re calling out is that Replit isn’t pricing based on actual value delivered.
Uber didn’t just randomly charge 15x the cost of a 2-minute ride; their pricing at least tied to distance, demand, and time. Here, a trivial correction or Q&A is being treated the same as a multi-minute orchestration run with sub-agents. That’s not “profit” — that’s a blunt minimum-fee model that punishes low-effort interactions and disincentivizes adoption.
Your drug dealer analogy also proves the opposite point: if Replit’s play is to lock people in and then hike costs arbitrarily, that’s not a “real cost of compute” — it’s extractive pricing once you’re trapped. And the community is right to push back before that becomes the norm.
This isn’t about wanting free AI or crying over profit margins. It’s about wanting transparent, proportional pricing that reflects the difference between a quick Q&A and a full-blown orchestration run. Without that, you’re not just “turning the screws,” you’re training your most engaged users to look for alternatives faster.
Yeah, again, I do sort of agree. My responses are, perhaps wrongly, aiming at the overall noise I’ve seen in the last week, across this platform, linkedin and reddit.
You say transparent, proportional pricing, and this is absolutely right. But also transparency in how the actual agent works - which is where agent 3 has gone massively wrong. My view is the product is the first issue to solve. If the product doesn’t work properly for the majority of your users, then there is no hope for transparency in pricing, because you don’t even know what you are setting prices for.
I’m just trying to see it from both sides. These companies all have massive shareholder pressure. And Replit know the other AI dev platforms will all be having the same conversations. So their thinking is probably “sure, go find another, but their prices will soon go up too”.
In fact, a real cynic would note these companies and their shareholders (many shared!) are all from a very small clique, and would think to how OPEC controls the global price of oil. You and I have no chance against this kind of cartel.
i have been trying to use Agent 3 since it was released and the performance is honestly dog crap, which is upsetting because I was getting far with Agen2 but it seems that every time they decide to just force an update it breaks everything, this is a very irresponsible way of running a company. There should always be a fallback option to go back until they have ironed out issues, what the hell is wrong with the team? I can’t even complain about pricing because it just hangs, it sits and does nothing for 10-15 minutes at a time and i’ve tired enabled the new modes and turning them off and everything, its just useless. I feel like I might have to export all my code and bounce if this doesn’t get resolved soon because after thousand and thousands of dollars i have spent on Replit, this is completely unacceptable for production, how do they expect to have professionals take it seriously if there is no consideration for stability on performance. To me it seems their servers are overloaded and they have not figured out how to distribute resources well, I have always noticed that the original Agent 2 just performed the best at like 1-3am and i could get so much done but as soon as the work hours start, it takes forever and is so slow. Now its not even responding at all, it loses context answering simple prompts and changes, and i’m not about to enable the high power model so they can stick me with $100 in charge for nothing. This is utterly unacceptable for the amount of money some of us are paying.
I am just starting to use agent 3. I agree that it takes longer to complete tasks and it is more expensive. but the question is am I getting more value for the longer processing time and potential automatic bug fixes ? I havent worked with it enough yet. I mean I would rather pay the same prices as before but then get these supposed added benefits.
I just asked it to increase the width of the autocomplete box on mobile device view for googles autocomplete api for addresses. It did increase the width on mobile and seems to have done it correctly, sort of (more width than i needed) but the cost was .77 cents. before a small change like that would have been .20 cents ?
so the thing is how do I know I am getting more value for such a simple change….
I’d disagree with any of us saying that Replit should be free to use after we pay for a license. The majority of us are used to “usage-based pricing” at this point. A lot of us have invested thousands into what we’re building. I’ve sat over the weekend in pure frustration that I haven’t progressed my app in 4 days because overnight prices skyrocketed with no warning, and no way to plan the future for our apps.
I’ve been fortunate to have been part of a few companies that we’ve been able to grow past $100M-$200M after joining early stage. For the company that was mid-market/enterprise, we’d communicate pricing changes nearly a year ahead of time so there were no surprises, even allowing customers to renew on legacy plans so they had the time to plan for new licensing metrics if they didn’t have ample time to plan. Heck, I think even at the time I left we had users on our legacy plans because we didn’t want to flip an SMB who had been using us for years on their heads and put them out of business based on the new licensing costs.
The second thing we’d do, is we’d work with the customer base to help show them how to optimize their spend, something we modeled after AWS, because they will go the extra mile to make sure customers weren’t spending an outrageous amount because they built something wrong.
We’d also have war room sessions, many time at like 7am (5am PST), to quickly communicate major issues like this with the community/our customers. I haven’t seen radio silence like we have seen since Agent 3 rolled out. It’s a little crazy that you’ve been the messenger on behalf of Replit. While we’re thankful for all you’ve done, not good enough for a 250 person company to not have official responses (granted, this all came in over the weekend so for the benefit of the doubt, I’d give at least 48-hours for this to happen).
After sitting in anger for 4 days, I feel incredibly fortunate that this happened now rather than later. I was reminded of my #1 rule of "Don’t build a company that is built on the back of another company.” It’s forced me to think through contingency plans and learn other tools, which I didn’t want to do, but am basically forced to do, and need to do so I’m not stuck when something like this happens*.*
