Jan 7, 2026 · 2 min read
I’ve blown through $50 credits of my $200 Cursor Ultra subscription in one day. I’ve never had so much fun with a model until Composer 1 (recently) became this good.
It’s nowhere near Claude Opus 4.5, don’t get me wrong. It makes very simple mistakes, but it seems to be on par with Claude Sonnet 4.5 (probably a bit less, still). However, it’s so amazingly fast. Whereas with Claude's models I’ll be waiting a bit every time, those times add up and cause you to weirdly lose the sense of “flow.” But when a model is up to 10 times as fast at generating, then you’re hardly ever out of it. It generates something super fast, I’ll read what it came up with, test if it actually works, and continue with the next task.
I think they got it this fast due to Anysphere’s partnership with Cerebras AI. I also see that Cerebras AI has its own Code Pro ($50, but with a token limit) and Code Max subscription ($200, larger token limit), and I immediately signed up for the waiting list for those. Perhaps I’ll be able to save costs with those subscriptions later on. I am a bit cautious about the reviews I’ve read on Reddit regarding its usage of the GLM models and their hallucination issues, so time will tell.
I’ve never had this much fun with a model. It seems like it hits the exact balance you want with prototyping (to quickly test, discard, and move on). It’s such a fun time to be a nerd.
Starting in 2026, I'll share more focused notes on product design, technology, and business. If you'd like them in your inbox, leave your email below. I'm always happy to connect via email, Bluesky, or LinkedIn (blergh).