Cursor 3.0 ships — multi-agent IDE replaces the Composer single-agent, $2.5B ARR run-rate clears Cognition+Windsurf
Cursor 3.0 shipped this week, replacing the Composer single-agent model with a native multi-agent runtime that supports concurrent agent execution on different tasks within a single workspace. Cursor disclosed a $2.5B ARR run-rate alongside the launch, putting the company at parity with Anthropic's Claude API revenue and ahead of Cognition+Windsurf's combined commercial line.
The multi-agent runtime is the structural product change. Composer's single-agent model required users to serialize tasks; Cursor 3.0's multi-agent runtime lets one agent run a long-horizon refactor while a second agent investigates a bug report from CI feedback and a third agent stages a documentation pass. The orchestrator is exposed as a new sidebar with explicit agent-state visualization.
The $2.5B ARR is the headline number that reframes the IDE-tools commercial tier. At $2.5B Cursor is past the GitHub-Copilot-2024 run-rate benchmark and into the tier where the IDE-tool category becomes a separately-strategic line item for enterprise procurement. Whether Cursor or Cognition+Windsurf wins the next round of enterprise consolidation depends largely on whether the multi-agent runtime or the Devin-bundled-quota model proves stickier in production.
Cursor — Cursor 3.0 launch announcement → · The Information — Cursor at $2.5B ARR → · Latent.Space — Cursor 3.0 multi-agent runtime →