Grok 4.5 first impressions
Cursor partnered with SpaceXAI to train Grok 4.5 and the model is now live. I just tried it inside Cursor and here are my first impressions:
It's not as good as Opus/Fable models when it comes to frontend design related tasks. I asked it to create a few webpages, and the UI is decent but not very good. I will compare its UI design capabilities with GPT-5.5, which is not very good.
Grok 4.5 is very fast when compared to the latest Opus or GPT models, even without the fast mode turned out. Also, it's reasonably priced at input $2 per million tokens and output at $6 per million tokens. As of now, I used around 13 million Grok 4.5 tokens inside Cursor, and it has costed me 2.6% of the monthly usage, as you see.
As you see above, Grok 4.3 billed inside Cursor as API pricing, but the new Grok 4.5 comes in the Auto + Composer pool.
Grok 4.5 has currently 2x limits inside Cursor for a week, and is completely free inside Grok Build if you have a paid plan for SuperGrok or X, as announced by the xAI team in this post.
Apart from this, the new model is 1.5T mixture of experts model, trained from scratch.
From my limited use, the model feels much better than Composer 2.5, and it feels like I am using GPT-5.5 inside Codex. It's fast and pretty good with non-frontend related tasks.
They mention in the blog post that:
Grok 4.5 and Composer 2.5 are two different model weight classes, and we're excited to support both sizes and weights. Composer 2.5 will remain offered, and we will release new models of this size going forward.
Loved this pricing comparison of Grok 4.5 with other popular models:
By the way, I am still experimenting, and will keep this post with new info I learn.
Webmentions