It’s fast. It seems to take instruction better than some of the other models. The responses it gives are detailed and quite good most of the time. Code quality leaves a lot to be desired though. Ridiculously complex code that needed breaking out into several helper methods.. Silly mistakes throughout… but on the whole.. not bad at all. Quite impressed with this model, to be fair.
1 Like
Agreed. Good. But I personally don’t think it is worth the price tag at $10 per Million tokens. $5.. maybe… with some definite caveats. 
1 Like
agreed. it is very pricey for what it is. what is your preferred model to use?
im kinda rolling with opus more most things right now.
Honestly, I direct my team to use GPT-5-mini. At $2 per Million tokens, with a base allowance for teams, they get 10 million tokens to work with before they kick into additional usage. 5-mini is fast and is a rational model so when its backed up by a good memory-bank, you can do quite a bit without costing a lot. Might take longer than a better model, but that can usually be managed with good context and a good prompt / plan.
I will use Composer 1 for mundane tasks, tasks that don’t require deeper analysis, or tasks that involve a lot of existing code/refactors.
I noticed early on that Composer 1 is NOT a thinking model. So it can handle a certain class of things, but without the ability to reason, there are definite limits. I’ve compared it to Grok Code 1 Thinking a lot, as both have similar speed (Grok is still faster, but Composer 1 is pretty fast), and the thinking Grok model usually wins.
I generally find myself using Sonnet 4.5 most of the time these days, though, as it just works better for the way I do things.
1 Like
Auto can do wonders if the code is modularized in small segments.
1 Like
composer is way behinf all others even grok works better sometimes the pricing is 3x more
1 Like
just switched to Kilo Code. Probably won’t come back to cursor.