Replit now uses Opus 4.6. This will make huge improvements to our apps.
But it is an expensive model. Does anyone know, are we likely to see our bills go up?
Replit now uses Opus 4.6. This will make huge improvements to our apps.
But it is an expensive model. Does anyone know, are we likely to see our bills go up?
Yes
, are you seeing any improvements in apps?
I can definitely see that my Flatrpicker issues that I’ve been struggling with for the last 2 days have been fixed after 5 tries with the new Opus 4.6, so I’m seeing improvements in that way.
I hope it doesn’t cost more than before the release, because it’s already daaamn expensive. ![]()
Not done any work in replit today so not sure. I will monitor next week and see.
But I can only assume Replit isn’t subsidising model rates. So unless they are being given huge discounts, then newer models will be pricier for us. Shame there isn’t a little more transparency on this.
Where can you see which model replit uses?
Sadly you can’t. You have to follow the replit team on X and hear about it that way.
Easy, just monitor your inbox for invoices ![]()
yeah very true indeed!!! ![]()
Ah, well, since I get invoiced daily, that should be no problem! : |
Thanks.
EDIT: Just reviewed all my usage and billing - can’t find the model anywhere. : (
It was a joke. Meaning you will see the invoice amounts shoot up if it is using opus 4.6.
Oh man, I wish it wasn’t a joke. But now that I know, that’s hilarious!
Good point.
Bills are up, mistakes are down, its a win win.
However, it cost me $1.40 to update the .md flies and .replit files for a project. It did a really good job, so do you complain about the price, or be happy with the result? I chose being happy with result, not cared about price if I get exactly what I wanted.
@grayhill5 I said to @Gipity-Steve the other day, seems around 20% higher with 4.6.
But the amount of work getting done per task seems higher. I also can’t remember the last time I had to correct something. It’s the accuracy now that pretty much seems to be 100%, except the odd silly thing, obviously.
It’s really hard to measure the cost / quality ratio.
The Agent does things, reviews with the architect, and makes iterative improvements within the same build cycle - less back and forth with the user. It seems to think better about dependency chains, and catches stuff itself. So probably does more work per single prompt.
That’s all helpful, but not free.
Maybe it consumes costs faster, because it does more autonomously, whereas previously maybe the same workload may have been handled over more days with back and forth with the user.