i tested whether composer 1.5 follows .cursorrules the same as other models. ran the same prompt with and without a branded types rule (asking it to use branded types for ID fields instead of plain strings).
result: with the rule it generated proper branded types (UserId, OrderId with __brand). without the rule it used plain string everywhere. same behavior i see with other models, so rule compliance doesn’t seem to have regressed.
speed-wise it does feel faster for routine stuff. haven’t tested it enough on complex multi-file tasks to say much there yet.
It’s in the Cursor application itself. Cursor Settings → Usage and Billing (or something like that). I forget what the section is called exactly, it does not show up if you’re on an enterprise/team plan, but does show up for me on my solo account.
I think the adoption rate might be a biased metric. I personally am using 1.5 half the time only because my API quota (which was suspiciously small this time and feels like it ran out with far fewer requests). Only 1.5 doesnt add $$ to my final payment. Which is what I suspect might be happening with others.
As for issues, I and others have stated various issues. It just fails to understand the task, understand the code and create incorrect code solutions 80% of the times (for me). It goes in loops and actually introduces incorrect logic that breaks other things. And yes I am giving it explicit guidelines and am using plan mode.
It’s just plainly a worse model. I personally do not trust any of the benchmarks (they can be fudged). I trust my own experience with developing with a model and right now that experience is terrible.
Composer 1.5 has been a huge upgrade for me, and the reason is kind of interesting.
With Composer 1, I didn’t always trust it to execute plans designed by Opus 4.6 — it would sometimes introduce subtle issues. So for tasks I wanted to be more careful with, I’d have Opus handle the execution instead.
With Composer 1.5, that’s completely changed. I now confidently hand off almost every plan to it for execution. That shift alone has been a massive productivity boost.
In cursor.com, I don’t have “Usage & Billing”. I have “Usage” and “Billing & Invoices”; none of them show me anything similar to what I am looking for.
This is my “Usage” section in the dashboard. And another thing about my dashboard, it doesn’t seem like I have access to this “Auto + composer” bucket. I have an “auto” bucket and a “composer 1.5” bucket. Maybe I am misunderstanding.
And in the cursor IDE, after setting the show usage option to always, I do get the text at the bottom that says the total percentage of usage. If I click on it, it shows the dollar amounts and the quota reset date
so cursor was working amazing for me, it understand the context quite well and almost everytime i dont need to explain same thing again and again for it to work.
but since today I am seeing that its hardly understanding the context and it doesnt feel smart anymore, isnt taking good decision.
I am always using it on AUTO mode and it works best for me.
Only thing that changes since yesterday is the update to Compose 1.5 agent.
Would it be effecting its reasoning? remember i am using it in AUTO mode.
Did anyone else notice same thing?
Is there a way i can use previous composer version please. this 1.5 is so bad.
