Cursor

Key Terms

• Agent mode - Cursor operates autonomously in the editor, reading files, running commands, and making multi-step edits. Available in the Agents Window (a dedicated panel). Source: Cursor – Agents Window
• Composer 2 - Cursor's in-house model, revealed by the community to be Kimi K2.5 with RL fine-tuning. Positioned as a fast, cost-effective option for agent tasks. Source: News – Item
• Bugbot - Cursor's automated code review tool. Reviews pull requests, suggests fixes, and can auto-fix issues. Available as a separate add-on at $40/user/mo. Source: Cursor – Bugbot
• Cloud agents - Agent sessions that run on Cursor's infrastructure rather than locally. Available on Pro and above. Source: Cursor – Agents
• FIM completion (Fill-in-the-Middle) - Cursor Tab, the inline code completion feature that predicts and inserts code at the cursor position using surrounding context. Source: Cursor – Tab
• On-demand usage - Pay-as-you-go model usage after the included plan amount is consumed, billed in arrears. Source: Cursor – Pricing

Latest Changes

First report for this supplier. All models, plans, and pricing are listed as current state.

• New model: Composer 2 (Kimi K2.5 with RL fine-tuning) revealed as Cursor's in-house model.
• Feature added: Cursor 3.2 (April 24): /multitask command, worktrees, multi-root workspaces.
• Feature added: Canvases (April 15): interactive visualizations in the Agents Window.
• Feature added: Cursor 3.1 (April 13): tiled layout, voice input.
• Feature added: Bugbot learned rules and MCP support (April 8). 78% resolution rate.
• Feature added: CLI debug mode and /btw command (April 14).
• Plan change: SpaceX secured option to acquire Cursor for $60B (April 21).

Plans

Individual Plans

Source: Cursor – Pricing

Business Plans

Source: Cursor – Pricing

Bugbot (Code Review)

Source: Cursor – Pricing

Note: Bugbot is a separate purchase from the IDE plan. Bugbot is not included in the $40/user/mo Teams IDE plan.

Usage Pricing

Every plan includes a set amount of model usage. After the included amount is consumed, on-demand usage is billed in arrears. Exact per-token rates are not published on the pricing page.

API Pricing

Cursor does not offer a standalone API. Usage is billed through the subscription plans described above. On-demand usage after included limits is billed in arrears at undisclosed per-token rates.

Model Performance / Benchmarks

Cursor does not publish independent benchmark scores for the Cursor IDE as a product. Cursor uses frontier models from Anthropic, OpenAI, and Google alongside its own Composer 2 model (Kimi K2.5 with RL fine-tuning). Benchmark scores for those underlying models are published by their respective providers. See the Anthropic, OpenAI, Google, and Moonshot AI raw reports for model-level benchmarks.

Bugbot code review resolution rate: 78%. Source: Cursor – 04 08 26

Latest News

Cursor 3.2: Multitask, Worktrees, Multi-root Workspaces (April 24, 2026)

• /multitask command runs async subagents to parallelize requests instead of queuing them
• Worktrees in the Agents Window for running isolated tasks across branches
• Multi-root workspaces allow a single agent session to target multiple folders for cross-repo changes

Source: Cursor – 04 24 26

Canvases (April 15, 2026)

• Interactive visualizations created by the agent (dashboards, tables, diagrams, charts)
• Durable artifacts in the Agents Window alongside terminal, browser, and source control

Source: Cursor – 04 15 26

CLI Debug Mode and /btw (April 14, 2026)

• /debug mode for finding root causes: generates hypotheses, adds log statements, uses runtime info
• /btw for side questions without derailing the main task
• /config interactive settings panel in CLI
• /statusline custom status bars

Source: Cursor – 04 14 26

Cursor 3.1: Tiled Layout and Voice Input (April 13, 2026)

• Tiled layout for running multiple agents in parallel panes
• Upgraded voice input with batch STT
• Branch selection in empty state, diff-to-file navigation, search filters

Source: Cursor – 3 1

Bugbot Learned Rules and MCP Support (April 8, 2026)

• Bugbot learns from PR feedback to create self-improving review rules
• MCP server support for additional context during code reviews
• Bugbot resolution rate: 78%
• "Fix All" action to apply multiple fixes in one operation

Source: Cursor – 04 08 26

SpaceX Acquisition Option (April 21-22, 2026)

SpaceX secured an option to acquire Cursor for $60 billion, with a $10 billion fallback for collaboration. Microsoft also looked at buying Cursor before the SpaceX deal (per CNBC). This was the dominant community narrative about Cursor this month.

Source: News – Item

Community Signals

Mass Migration to Claude Code

The most significant community trend this month is developers switching from Cursor to Claude Code.

• Reddit: "Anyone switched from Cursor to Claude Code?" (114 upvotes, 130 comments). Source: Old – Top
• Reddit: "Aaaaand I cancelled my Cursor subscription" (179 upvotes, 153 comments). Source: Old – Top
• HN user spent $1,800/month on Cursor, then $200/month after switching to Claude. Source: News – Item
• Enterprise user spending ~$2k/week on Cursor with premium models, switched to Claude Code Max for 1/10th the price. Source: News – Item
• "My company is actively encouraging our devs to NOT use Cursor because of how much more expensive it is compared to other tools." Source: News – Item

Cursor Agent is Claude Code SDK Behind a Proxy

• Community analysis revealed Cursor Agent is Anthropic's Claude Code SDK running behind a local HTTP proxy. Reddit: 190 upvotes, 54 comments. Source: Old – Top
• HN: "The interesting question is what Cursor adds on top - their UX and context management decisions are where differentiation actually lives." Source: News – Item

Hidden Cache Token Billing

• User on 2x Ultra plans: usage jumped from ~$60-100/month to $500+ in days (projected ~$1,600/month). Root cause: Cursor builds a large hidden prompt state that is prompt-cached via Anthropic's API. One concrete example: user input ~4k tokens, cache read tokens ~21 million, one call cost ~$12. UI shows "max 200k context" but billing reflects much larger hidden history. Source: News – Item

Teams Plan Billing Issues

• Team added a seat and removed it within seconds. Cursor billed $450 for a full annual seat with no UI warning. Teams plan at $40/user/mo only includes $20 of AI usage credits (remaining $20 is for "team features"). Source: News – Item

Cursor V3 Regression

• Reddit: "Cursor V3 is a significant regression" (213 upvotes, 118 comments). Source: Old – Top

Composer 2 Revealed as Kimi K2.5

• Community discovered Cursor's "in-house model" Composer 2 is actually Kimi K2.5 with RL fine-tuning. Cursor confirmed via co-founder Lee Rob on Twitter. HN: 276 points, 168 comments. Source: News – Item
• User feedback: Composer 2 is "not as intelligent" but has "as good as or better intuition" with "way better pricing" and is "about 100x faster" than Opus.

SpaceX Deal Privacy Concerns

• HN: "Cursor alternative, EU-based or privacy-focused" citing "the recent SpaceX deal, and how many privacy-related bugs they've shipped recently that they can't seem to fix, like not being able to delete chat history." Source: News – Item
• Reddit: "Did Elon just kill the appeal of Cursor?" (212 upvotes). "So Elon Musk will end up having the source code of millions of developers that use Cursor?" (134 upvotes). Source: Old – Top

Cursor 3 Launch Reception

• HN: 544 points, 406 comments. Source: News – Item
• Agent swarms skepticism: senior engineers prefer serial one-task-at-a-time workflow
• "Vibe-first" direction criticism: "I wish they'd keep the old philosophy of letting the developer drive and the agent assist"
• "Cursor's competitive window closing" sentiment expressed by multiple commenters

Enterprise Readiness

Transparency Gaps