Cursor

Executive Summary

What it is: Cursor by Anysphere is a multi-model AI coding IDE (fork of VS Code) with agent mode, cloud agents, code review (Bugbot), CLI, and marketplace. Individual plans are now tiered: Pro ($20/mo, $20 API usage included), Pro+ ($60/mo, $70 included), and Ultra ($200/mo, $400 included). Teams is $40/user/mo with usage-based billing. Enterprise is custom pricing with pooled usage. Cursor supports 20+ models from Anthropic, OpenAI, Google, xAI, and Moonshot, plus its proprietary Composer models.

What to watch out for: Cursor's API credits for frontier models (Opus, GPT-5.5) burn significantly faster than the same models in other harnesses. Users report Sonnet 4.6 consuming 7-9% of usage per prompt vs 1% in VS Code with Claude directly. A side-by-side comparison found Claude Code is ~5x cheaper overall and ~38x cheaper on frontier models at the same $200/mo price point. The SpaceXAI partnership to train a larger Composer model signals a shift toward proprietary models, which could lock users into Cursor's ecosystem. A US Congressional probe into Anysphere's use of Chinese AI models adds regulatory risk.

Bottom line: Cursor's IDE experience and Composer 2.5 model make it the strongest harness for developers who want a polished, integrated coding agent. However, the cost-value equation is under severe pressure. At $200/mo, Cursor Ultra provides roughly 18 agent-hours of frontier model usage, compared to 678 hours with Claude Code Max 20x at the same price. Composer 2.5 mitigates this by offering near-frontier quality at much lower cost, but developers who need Opus or GPT-5.5 should budget for significant overage charges.

Key Terms

  • Composer - Cursor's proprietary coding model family. Composer 2.5 (launched May 18) is built on Moonshot's Kimi K2.5 checkpoint with additional training. Standard pricing: $0.50/$2.50 per MTok. Fast (default) pricing: $3/$15 per MTok. Source: Cursor – Composer 2 5
  • Auto - Cursor's agent mode that uses Composer models with automatic tool use. Auto requests are exempt from the Cursor Token Rate surcharge. Source: Cursor – Pricing
  • Cursor Token Rate - A $0.25/MTok surcharge on top of model API pricing for non-Auto agent requests on Teams plans. Source: Cursor – Pricing
  • Bugbot - Cursor's agentic code review tool. Runs on usage-based billing with three effort levels (Default, High, Custom). Source: Cursor – 05 11 26
  • Cloud agents - Hosted coding agents that run on Cursor's infrastructure. Support multi-repo environments, Dockerfile configuration, and automations. Source: Cursor – Cloud
  • Max Mode - Required to use the most expensive models (GPT-5.5, GPT-5.3-Codex). Source: Cursor

Latest Changes

Changes since the 2026-04 report.

  • Plan change: Individual plans now tiered into Pro ($20/mo), Pro+ ($60/mo), and Ultra ($200/mo) with different included API usage amounts ($20, $70, $400 respectively). Previous plan was a single tier at $20/mo. Source: Cursor – Pricing
  • New model: Composer 2.5 launched May 18, built on Moonshot's Kimi K2.5 checkpoint. Standard: $0.50/$2.50 per MTok. Fast (default): $3/$15 per MTok. Trained with 25x more synthetic tasks than Composer 2. Double usage included for first week. Source: Cursor – Composer 2 5
  • New model: Claude Opus 4.8 available (following Anthropic's May 28 launch).
  • New model: GPT-5.5, GPT-5.4/5.4 Mini/5.4 Nano, Gemini 3.5 Flash, Grok 4.20, Grok Build 0.1, Kimi K2.5 all available. Source: cursor.com/pricing docs
  • Partnership: SpaceXAI partnership announced May 18 to train a "significantly larger model from scratch" using 10x more compute on Colossus 2 (million H100-equivalents). Source: Cursor – Composer 2 5
  • Feature added: Auto-review run mode (v3.6, May 29). Shell, MCP, and Fetch calls routed through allowlists, sandboxes, or a classifier subagent. Source: Cursor – Auto Review
  • Feature added: Shared canvases (v3.5, May 20). Share interactive agent artifacts via link. Source: Cursor – Shared Canvases
  • Feature added: /loop skill (v3.5, May 20). Run a prompt repeatedly on schedule or until outcome achieved. Source: Cursor – Shared Canvases
  • Feature added: Cursor in Jira (May 19). Assign work items or mention @Cursor in comments. Source: Cursor – 05 19 26
  • Feature added: Cursor in Microsoft Teams (May 11). Mention @Cursor in channels to delegate tasks. Source: Cursor – Microsoft Teams
  • Feature added: Development environments for cloud agents (v3.4, May 13). Multi-repo, Dockerfile-based, 70% faster cache hits. Source: Cursor – 05 13 26
  • Feature added: Multi-repo and no-repo automations (v3.5, May 20). 5 new templates for non-code tasks. Source: Cursor – 05 20 26
  • Feature added: Bugbot effort levels (May 11). Default (0.7 bugs/run, 79% resolved at merge), High (0.95 bugs/run), Custom. Source: Cursor – 05 11 26
  • Feature added: Build in Parallel and Split PRs (v3.3, May 7). Execute plan tasks simultaneously via async subagents. Source: Cursor – 05 07 26
  • Feature added: Cursor TypeScript SDK (April 29-30). Build programmatic agents using Cursor's infrastructure. Source: Cursor – Typescript Sdk

Plans

PlanPrice (monthly)API Usage IncludedKey Inclusions
Hobby$0NoneLimited agent requests, limited tab completions, no credit card required
Pro$20/mo$20Extended agent limits, frontier model access, MCPs/skills/hooks, cloud agents, Bugbot on usage-based billing
Pro+$60/mo$70Everything in Pro with more included usage
Ultra$200/mo$400Everything in Pro+ with maximum included usage
Teams$40/user/moUsage-basedCentralized billing, team marketplace, Bugbot, cloud agents with shared context, usage analytics, privacy mode, SAML/OIDC SSO
EnterpriseCustomPooledEverything in Teams + pooled usage, invoice/PO billing, SCIM, repository/model/MCP access controls, audit logs, AI code tracking API, priority support

Pricing notes:

  • On-demand usage available when included amount is consumed, billed at same API rates
  • Teams plans include $0.25/MTok Cursor Token Rate on top of model pricing for non-Auto requests
  • Auto requests are exempt from Cursor Token Rate

Source: Cursor – Pricing, Cursor – Pricing

API Pricing

Cursor passes through model API rates. Key models available:

ModelInput ($/MTok)Output ($/MTok)Notes
Composer 2.5 (standard)$0.50$2.50Proprietary, best value
Composer 2.5 (fast)$3.00$15.00Default mode
Claude Haiku 4.5$1.00$5.00
Claude Sonnet 4.6$3.00$15.00
Claude Opus 4.8$5.00$25.00Fast mode available
GPT-5 mini~$0.20~$1.00
GPT-5.4$2.50$15.0090% cached input discount
GPT-5.4 Mini$0.75$4.50
GPT-5.4 Nano$0.20$1.25
GPT-5.3-Codex$1.75$14.00Requires Max Mode
GPT-5.5$5.00$30.00Requires Max Mode
Gemini 3.5 Flash$1.50$9.00
Gemini 2.5 Pro~$1.25~$10.00
Grok 4.20$2.00$6.002x cost over 200k tokens
Grok Build 0.1$1.00$2.00New
Kimi K2.5$0.60$3.00New

Source: Cursor – Pricing

Model Performance / Benchmarks

Cursor does not publish standardized benchmarks for Composer models. From the Composer 2.5 blog post:

  • Trained with 25x more synthetic tasks than Composer 2
  • "Substantial improvement over Composer 2 in sustained work, instruction following, and collaboration"
  • New training techniques: targeted RL with textual feedback, sharded Muon, dual mesh HSDP

Cursor CEO Michael Truell stated in the Opus 4.8 announcement that "On CursorBench, Claude Opus 4.8 exceeds prior Opus models across every effort level" and that "Tool calling is meaningfully more efficient, using fewer steps for the same intelligence."

Source: Cursor – Composer 2 5, Anthropic – Claude Opus 4 8

Latest News

Composer 2.5 Launch and SpaceXAI Partnership (May 18, 2026)

Cursor launched Composer 2.5, a proprietary coding model built on Moonshot's Kimi K2.5 checkpoint. Also announced a partnership with SpaceXAI to train a significantly larger model from scratch using 10x more compute on Colossus 2 (million H100-equivalents).

Source: Cursor – Composer 2 5

SpaceX Option to Acquire Cursor for $60B (April 21, reported into May)

SpaceX announced it has an option to acquire Cursor for $60B, or alternatively pay $10B for a partnership. Covered by Reuters, NYT, Bloomberg, CNBC. HN: 989 comments, 823 points.

Source: News – Item

$300M ARR and $50B Valuation (April-May 2026)

Cursor in talks to raise $2B+ at $50B+ valuation. Lenny's Newsletter profiled Cursor as "The $300M ARR AI tool that engineers can't stop using."

Source: TechCrunch – Sources Cursor In Talks To Raise 2B At 50B Valuation As Enterprise Growth Surges

US Congressional Probe into Chinese AI Model Use (April 29-30, ongoing)

US House panels are probing Anysphere over use of Chinese AI models, which could affect Cursor's ability to use models like DeepSeek.

Source: Nextgov – House Panels Probe Airbnb Anysphere Over Use Chinese Ai Models

PocketOS Production Database Deletion (April 25-27, discussed into May)

A Cursor agent running Claude deleted PocketOS's entire production database and backups on Railway in 9 seconds. Led to creation of MCP Defender and BetterClaw safety tools.

Source: News – Item

Cursor SOC 2 Certified

Noted on website footer. Enterprise security validation.

Source: Cursor – Security

Community Signals

Pricing Comparison: Claude Code ~5x Cheaper

Andrew Shu published a side-by-side comparison (12 experiments at $200/month):

Tool + PlanAgent-Hours/Monthvs. Cursor Ultra
Cursor Ultra ($200)~138 hours1x
Codex Pro ($200)~220 hours~1.6x
Claude Code Max 20x ($200)~678 hours~4.9x

Frontier-only (Opus, GPT-5.5): Cursor Ultra provides ~18 agent-hours vs Claude Code's ~678 (38x difference).

Cursor's usage is split 87% Auto/Composer (cheap) and 13% API credits (frontier models, expensive).

Source: Ashu – Claude Code Vs Cursor Pricing

Community Pricing Sentiment (Reddit, May 31)

  • troubleshootmertr (11 upvotes): "Cursor is probably the best bang for buck coding plan by far. Auto and composer 2.5 are great and dirt cheap. I pay $20 a month."
  • CoreDirt (4 upvotes): "$60 cursor and $100 Claude and $20 codex. I use them all within cursor. I had the $200 cursor ultra for a while but if you're hammering the opus or sonnet you will burn through it."
  • UnderstandingDry1256 (5 upvotes): "2x Ultra accounts, absolutely worth it. I don't care using stupid models like composer or anything worse than the latest Opus or GPT."
  • camlp580 (2 upvotes): "Just left cursor actually. I was already paying for Claude and GPT for other needs. Made more sense for me to pull them into VS Code."
  • andupotorac (4 upvotes): "Paying around 250eur, it's producing around 50K eur."
  • Enterprise user adelope: "On my 9-5 on the other hand, i use ~$10k/m"

Source: Reddit – Be Honest What Are You Actually Paying For Cursor

API Usage Burns Faster in Cursor

Reddit user reports: "On vs code, if I select Sonnet 4.6 the 'usage' API bar barely moves, about %1. On cursor, for the same model it moves around %7 or %9. After 3 chat prompts already at %20."

Source: Reddit – My Api Usage Goes Down Too Fast I Am Doing

Enterprise Adoption Signals

  • $300M ARR and 140,000+ organizations
  • Users reporting $1,000+/month spend (almost entirely on Composer)
  • Enterprise user reporting $10k/month spend
  • Cursor CEO warning about vibe coding building "shaky foundations"

Safety Incidents

PocketOS production database deletion by a Cursor agent led to community building safety tools: MCP Defender (OSS AI firewall for MCP), BetterClaw (compiles workflow descriptions into directed graphs to gate agent tools), and Clampd-Guard.

Source: News – Item

Enterprise Readiness

FeatureAvailable?Details
SSO (SAML)YesTeams and Enterprise plans. Source: Cursor – Pricing
SSO (OIDC)YesTeams and Enterprise plans. Source: Cursor – Pricing
SCIMYesEnterprise plan only. Source: Cursor – Pricing
Audit logsYesEnterprise plan only. Source: Cursor – Pricing
IP indemnityNoNot mentioned on pricing, enterprise, or security pages.
Data residencyNoNot mentioned. Cloud agents run on Cursor's infrastructure.
HIPAANoNot mentioned on any page.
Air-gapped / on-premNoCursor requires internet connectivity.
SLANoNo publicly documented SLA.
Admin controls (RBAC)YesRepository, model, and MCP access controls. Auto-run, browser, and network controls. Enterprise plan. Source: Cursor – Pricing
SOC 2YesSOC 2 Certified. Source: Cursor – Security

Transparency Gaps

GapDetailsSeverity
Exact usage limits per planPro/Pro+/Ultra list "$20/$70/$400 included" but do not specify how many agent requests, hours, or tokens this translates to. "Generous included usage" for Auto/Composer is not quantified.High
Cursor Token Rate scopeThe $0.25/MTok surcharge on Teams plans applies to "non-Auto agent requests" but the full list of affected features is not clearly documented.Medium
API usage consumption rateUsers report Cursor consuming 7-9x more usage per prompt than VS Code with the same model. Whether this is due to additional context, system prompts, or Cursor's infrastructure overhead is unclear.High
Composer 2.5 benchmarksNo standardized benchmark scores published. Claims of "substantial improvement" are not quantified against competitors.Medium
SpaceXAI model timelineThe partnership to train a "significantly larger model" has no published timeline, pricing, or feature specifications.Medium
US Congressional probe outcomeThe investigation into Anysphere's use of Chinese AI models is ongoing. Impact on Cursor's model availability is uncertain.Medium