Cursor introduced Composer, its first claimed competitive coding model, alongside Cursor 2.0, an IDE modeled after Visual Studio Code. Composer, built with reinforcement learning and a mixture-of-experts architecture, is pitched as a frontier model “4x faster than similarly intelligent models.” Cursor’s internal benchmarks suggest it trails the best frontier models in intelligence, tops open and speed-oriented rivals, and far outpaces them in speed. A new multi-agent interface lets developers run many agents in parallel via git worktrees or remote machines to compare results. Early reactions in a small sample cite high cost and a perceived capability gap.
Reviewed by JQJO team
#coding #ide #software #developer #innovation
Comments