Claude Opus 4.8 is now available in Cursor!
Opus 4.8 is now available in Cursor. On CursorBench, it’s able to work much more efficiently than Opus 4.7. We’ve also found it to be more persistent on harder tasks.
Have you tried it yet?
Could we see the cursor bench output?
If possible, I suggest you directly remove Cursor’s non-max mode. After all, all future models will force users to use max mode anyway. Don’t you agree?
Still max mode? As 4.7 still have normal context size on 300k. Why don’t user can select default context size as 300k and not going to use max mode?
Available now on Cursor · CursorBench!
It’s too expensive…I won’t even try it.
10x more expensive than Composer 2.5
2x more expensive than my go-to model GPT 5.4
I wonder if when these new models come out Anthropic/OpenAI start to throttle the performance of older models gradually. This makes the newer models appear even better and can push people out of older models and into new models, even if the end result performance is the same eventually. Not saying this is going on, but it seems like an indirect way to increase the cost of models without every actually increasing the cost of the current model you are using, which people would be more likely to object.
I’ve seen similar posts on X about Anthropic, but GPT-5.4 still works great in both Cursor and Codex. GPT-5.5 outperforms it on boundary value problems, but that’s expected.
But… from mine side it think a lot of better, give a choices for me, when stuck or somethink… and i feel better responses…than Composer 2.5
No 50% discount during launch period this time?
I think Cursor is past the discount phase. All the companies are done losing money lol.
I don’t see this on mine.
The project’s bug tracking is done in BUGS.md, not on GitHub. I suggest Opus pick any one section and close all the issues in it.
I knew he’d pick the only section with only one open issue 
Sweet summer child.
The 50% discount was enough to get me to try out GPT 5.5 and now I’ve switched to that after having been on the Anthropic models for the past several months. Unless there’s an incentive like a discount, I’ll probably stick with what I know works, rather waste tokens experimenting with something new.
The companies are losing customers and not gaining new ones by not having occasional discount periods.