Cursor doubles Composer limits

Cursor said it temporarily doubled Composer 2 usage limits for the weekend, easing the hourly caps on its in-editor coding agent. (x.com) The change applies to Composer 2, Cursor’s first-party model for “the agent loop,” the part of the editor that can plan, edit files, and take many actions across a coding task. Cursor announced Composer 2 on March 19, 2026, and described it as a model built for long, multi-step software work. (cursor.com 1) (cursor.com 2) Cursor’s own announcement said Composer 2 can handle tasks requiring “hundreds of actions,” and the company sells it in standard and fast modes priced at $0.50 per million input tokens and $2.50 per million output tokens, or $1.50 and $7.50 for the faster version. On individual plans, Cursor says Composer usage comes from a separate usage pool with included capacity. (cursor.com 1) (cursor.com 2) An agent-based coding workflow is software that does more than answer a prompt once: it reads code, makes a plan, edits files, runs commands, and loops until a task is done. That makes usage limits more visible, because a single request can trigger many steps instead of one chat reply. (cursor.com 1) (cursor.com 2) Cursor’s help pages say usage is governed by plan budgets and reset windows, and the company has been steering users toward clearer pools for model use and on-demand spending. A temporary limit increase fits that model by giving users more room to test longer runs without changing the underlying pricing structure. (cursor.com) (cursor.com) The timing also lines up with Cursor’s push to make Composer 2 the default engine for harder coding tasks. In its launch post, the company said the faster Composer 2 variant is now the default and positioned the model as a lower-cost option for agentic software engineering. (cursor.com) (cursor.com) Cursor framed Composer 2 as stronger than earlier versions on coding benchmarks including Terminal-Bench 2.0 and SWE-bench Multilingual, with scores of 61.7 and 73.7 in its March 19 report. Those benchmarks are designed to test longer, tool-using coding sessions rather than short autocomplete snippets. (cursor.com) (cursor.com) For users, the weekend change is a short-term invitation to run bigger jobs inside Cursor while the company tries to get more developers comfortable handing longer coding loops to an agent. (x.com)