cheetah unmasked: Composer is here

In my last post I wrote about the mysterious model “cheetah” – including speculation that it might be Gemini 3.0. Today the mystery is officially solved: cheetah was not Gemini, but an in-house development by Cursor. With Cursor 2.0, the model has now been officially introduced as Composer (read the announcement ).

What is Composer – the short version

Composer is an AI model built explicitly for software development. The focus is less on eloquent prose and more on: “Deliver good code quickly in real codebases.” According to Cursor, Composer is noticeably faster (thanks in part to lean context usage) and was trained on real-world challenges from large repos – including tool support like semantic search to pinpoint and modify relevant code (details ). In practice that means: suggestions come fast and are often commit-ready straight away.

What else does Cursor 2.0 bring?

Alongside the model announcement, Cursor is introducing a multi-agent interface: multiple agents can work simultaneously on a task – separated into Git worktrees or even running on remote machines. At the end I pick the best solution or merge the strongest parts together. For more complex changes in particular, this significantly improves the hit rate (more here ).

My takeaway after the cheetah chapter

Looking back, the feeling I had with “cheetah” fits the explanation perfectly: pragmatic, fast, code-focused. Composer feels like the logical, “official” continuation – with a proper name, roadmap, and product around it.

What I’ll be paying close attention to going forward:

Personally, I’m glad the “cheetah mystery” has been neatly resolved. The move toward Composer + multi-agents feels like a genuine product moment, not just another model in a dropdown. I’ll keep at it, put Composer up against real challenges in day-to-day work – and report back on where it shines and where it still creaks.

– further reading: Composer announcement  · Cursor 2.0  · my earlier post: cheetah

Legal notice | | My GitHubRSS