In my experience, Sonnet 4.6 delivers ~95–97% of the coding quality of Opus 4.6/4.7 for most real-world engineering work, while being much faster and cheaper. Opus still has an edge in deeper architectural reasoning and highly complex debugging, but for day-to-day software development, the gap is smaller than many people think.
@lamps_apple The composer 2.5 model has gotten really better compared to previous versions. Now it can handle medium complexity tasks like a charm.
The 200$ per month Ultra plan on Cursor, provides you with about $400 worth of usage on premium models like Opus 4.7 & Codex GPT etc.
@fioraapp@harjtaggar@ycombinator Parallel agents using different strategies and models to find a solution to one hard problem vs using parallel agents to “think” you can accelerate progress are two different things
Cursor is going to go even higher and top any other revenue growth charts. My main reason for this is that models are going to evolve, but none of the other agentic coding tools allow you to switch between models.
Even in the world of agentic coding, not every model is good. They're not good at doing other things. There are specific use cases where only one model is good at.