GPT-56T 861 —
MUSE-SPK 837 —
GPT-56SC 789 -0.1%
GLM-5 781 —
CL-OP55X 779 -0.1%
GROK-46H 779 -0.1%
QWEN-38X 748 —
GPT-6A 743 —
KIMI-K3X 742 —
CL-FAB5H 697 -0.1%
CL-OP5H 674 -0.1%
GEM-38FH 672 —
CL-OP5X 669 -0.1%
CL-OP55H 667 -0.1%
CL-OP46H 656 -0.2%
CL-OP47H 647 -0.2%
GPT-56S 617 -0.2%
GEM-37FH 609 -0.2%
GEM-36FH 592 -0.2%
CL-OP48H 587 -0.2%
CL-OP47 580 -0.2%
GEM-35FH 579 -0.2%
GPT-55H 540 -0.2%
INKL 531 —
GEM-31P 511 -0.2%
CL-OP46 498 —
GEM-3P 498 —
CL-OP48 492 —
GPT-52 464 —
GPT-55 423 —
GPT-56T 861 —
MUSE-SPK 837 —
GPT-56SC 789 -0.1%
GLM-5 781 —
CL-OP55X 779 -0.1%
GROK-46H 779 -0.1%
QWEN-38X 748 —
GPT-6A 743 —
KIMI-K3X 742 —
CL-FAB5H 697 -0.1%
CL-OP5H 674 -0.1%
GEM-38FH 672 —
CL-OP5X 669 -0.1%
CL-OP55H 667 -0.1%
CL-OP46H 656 -0.2%
CL-OP47H 647 -0.2%
GPT-56S 617 -0.2%
GEM-37FH 609 -0.2%
GEM-36FH 592 -0.2%
CL-OP48H 587 -0.2%
CL-OP47 580 -0.2%
GEM-35FH 579 -0.2%
GPT-55H 540 -0.2%
INKL 531 —
GEM-31P 511 -0.2%
CL-OP46 498 —
GEM-3P 498 —
CL-OP48 492 —
GPT-52 464 —
GPT-55 423 —
← Back to feed

Cursor 3 Rebuilds the IDE from Scratch Around Agents — Composer 2 Beats Claude Opus 4.6 on Terminal-Bench

Cursor launched Cursor 3 on April 2, its most significant product shift since it forked VS Code. The new release introduces the Agents Window — a standalone interface built from scratch to orchestrate fleets of AI coding agents across local machines, cloud sandboxes, and remote environments simultaneously.

The traditional IDE editor remains available alongside the new interface, accessible via Cmd+Shift+P → Agents Window. Cursor frames this as the “third era” of AI-assisted development: after tab autocomplete and synchronous agents, parallel autonomous development at scale.

What Changed

Agents Window replaces the Composer pane as the primary agent surface. All agents — whether kicked off from desktop, mobile, web, Slack, GitHub, or Linear — appear in a unified sidebar. More than 30% of the PRs Cursor merges internally are now opened by agents running in cloud sandboxes.

Cloud-local handoff allows seamless session transfer. Start a task locally, move it to the cloud when you close the laptop, pull it back for local testing. The handoff is one action in each direction.

Design Mode lets developers click and annotate UI elements in an integrated browser to give agents precise visual targeting — eliminating the round-trips of describing interface changes in text.

Best-of-N uses worktrees to run the same prompt across multiple models or configurations simultaneously, compare outputs side by side, and select the strongest result.

Automations (shipped in early March) add always-on agents that trigger from Slack messages, Linear issues, GitHub events, PagerDuty alerts, and custom webhooks. Agents build memory across runs and can identify patterns over time.

Composer 2 Benchmark Performance

Cursor’s proprietary Composer 2 model, launched March 19, underpins the Agents Window.

ModelSWE-bench ProTerminal-Bench 2.0HumanEval+
Composer 261.361.773.7
Claude Opus 4.6—58.0—
GPT-5.4—75.1—

Token costs dropped approximately 86% compared to Composer 1.5. Pro plan pricing remains $20/month. Cursor says Composer 2 is “great for iterating quickly” — the positioning targets the large middle band of coding tasks where Claude Opus 4.6 is overkill on cost and GPT-5.4 is overkill on latency.

The IDE vs Terminal Debate

Cursor 3 is the strongest version of the IDE-centric argument in agentic coding. Claude Code, which holds the commanding lead in terminal-native workflows, represents the opposing thesis: agents belong at the shell, not inside a windowed editor.

The Cursor bet is that most professional developers won’t abandon LSP integration, a browser pane, and a visual diff interface even as agents write more of the code. The terminal-native camp disagrees. Cursor 3 does not resolve this debate; it sharpens the stakes.

The Agents Window is available now to all subscribers. Enterprise teams get new admin controls for security and attribution settings. The Cursor Marketplace launched with 30+ integrations from Atlassian, Datadog, and GitLab.