Cursor

Anysphere's AI-native code editor — packaging Claude / GPT into a coding IDE, becoming a phenomenal SaaS that ran from dark horse to $1B+ ARR in 2025.

1. Core Product / Service

Cursor is a desktop IDE forked from VS Code, with all features redesigned around "AI helps you write code":

• Tab autocomplete — multi-line / multi-file predictive autocomplete, with cross-file context capture
• Cmd+K edit — natural-language instructions to modify current file / selected code blocks
• Composer / Agent — multi-file-level agent tasks ("change all useState to useReducer"), cross-file plan + auto apply
• Chat / Ask — sidebar conversation, bound to entire codebase as context
• Codebase indexing — background embed + RAG of project so the model can see your codebase
• Background agents (launched 2025) — long-running background tasks, PR-level agents

Model layer is fully abstracted: users select Claude Sonnet 4.7 / Claude Opus 4.7 / GPT-5 / Gemini 3 Pro / DeepSeek v4 / Cursor's self-trained cursor-small in settings. Backend Cursor often pulls tokens via openrouter / directly connects claude-ai / OpenAI / Google AI Studio.

Pricing (2026-05):

• Hobby Free — rate-limited GPT-4o-mini / Cursor Tab, limited Composer
• Pro $20/month — ~500 advanced model requests/month (Claude / GPT-5) + unlimited cursor-small + Tab unlimited
• Pro+ $40/month — 1500 requests / month (added late 2025)
• Ultra $200/month — heavy tier (added late 2025)
• Business $40/seat·month — SSO, SCIM, no-training commitment, centralized usage billing
• Enterprise — negotiated pricing

2. Target Users & Pain Points

• Individual developers / independent engineers: originally using VS Code + Copilot $10, found Cursor Tab significantly more accurate (because it sees cross-file context), willing to pay $20.
• Small teams / startups: bulk-purchase Cursor Business as productivity tool, $40/month/person saves senior devs 20-40% coding time — high return.
• Large codebases / context-important projects: Cursor indexes the entire repo, a tier stronger than Copilot's single-file context.
• Heavy agent users: Composer / Agent mode turns "modify 50 files" into a 5-minute task.

3. Competitive Landscape

Cursor is clearly leading in the "individual / small team IDE experience" track, capturing in 2025 most Silicon Valley startup engineer subscriptions + many FAANG personal subscriptions (when Business is blocked by enterprise IT policy).

4. Unique Observations

• $20 / $40 / $200 tiers reflect explosion of model costs: in 2024 when Cursor was single-tier $20, Claude Sonnet 3.5 was still cheap and running Composer hundreds of times per month was fine. In 2025 with Sonnet 4 / 4.5 / 4.7 inference time and token volume doubled, Cursor had to raise tiers continuously (first adding $40 Pro+, then $200 Ultra) — this is a live sample of the framework.md L3 price-rise → L4 must-rise dynamic.
• Astonishing token consumption: Cursor heavy users (Composer + Agent fully on) consume 50-200M tokens/month, 50-100x of ChatGPT Plus users. This explains why Cursor had to launch Ultra $200 tier: a single Pro $20 user running Sonnet 4 at full tilt for a month, raw API cost can exceed $100, Cursor must lose money.
• Tight implied gross margin: industry estimates Cursor overall gross margin in 30-50% range (far below typical SaaS 70-80%), because underlying tokens are extremely variable costs. Cursor must use cursor-small / self-trained models to push down "low-value tasks" (Tab autocomplete) costs — this is the key economic motivation for its self-trained models.
• ARR running extremely fast: 2024-04 ARR ~$1M; 2024-09 ~$100M; 2025-05 ~$500M; late 2025 industry rumors ARR breaking $1B (The Information etc.). From $1M to $1B in less than two years — a rare curve in SaaS history.
• Distribution moat = IDE replacement = workflow ownership: unlike Copilot as a VS Code plugin, Cursor makes users replace the entire IDE. Once an engineer makes Cursor their main IDE, migration cost (shortcuts, settings, index data) is extremely high. This is the framework.md L4 Path 3 "deepest lock-in" sample.
• Data / workflow lock-in from codebase index: Cursor builds codebase embedding locally (Pro) + cloud (Business+), and after long use "it understands this project of yours." Switching to Copilot means losing this context.
• Strategic risk: claude-ai released Claude Code (CLI) late 2025 + Anthropic's own IDE integration is being called for — Cursor's relationship with its largest model supplier is delicate, as Anthropic is both supplier and potential competitor. Same story playing out with OpenAI / Microsoft pushing Copilot Workspace.
• Impact of OpenAI acquiring Windsurf (late 2025 rumored / partly confirmed): OpenAI's strategic intent to own IDE entry is a structural threat to Cursor — see windsurf entry.

5. Financials / Funding

Parent Anysphere:

• Series A $8M (2023-10, OpenAI Startup Fund led)
• Series B $60M (2024-08, Andreessen Horowitz led, valuation ~$400M)
• Series C $105M (2024-12, Thrive Capital led, valuation ~$2.6B)
• Series D $900M (2025-06, Thrive + a16z + Accel, valuation ~$10B)
• Series E (late 2025 rumored, valuation $20-30B range, not fully public confirmed)
• ARR: late 2025 industry estimate $1B+

6. People & Relationships

• Founders: Michael Truell (CEO), Sualeh Asif, Arvid Lunnemark, Aman Sanger — four engineers in their 20s from MIT
• Investors: a16z, Thrive, Accel, OpenAI Startup Fund, Benchmark
• Underlying models: claude-ai (Sonnet / Opus, mainstay), chatgpt (GPT-5 / o3-pro), gemini (Gemini 3 Pro), DeepSeek, cursor-small (self-trained)
• Competitors: github-copilot, windsurf, Claude Code, devin, replit
• Ecosystem partners: Anthropic (most important model supplier), OpenAI, Google AI Studio
• Risk relationship: model suppliers = potential competitors

Sources

• https://cursor.com/pricing (2026-05-10)
• https://www.theinformation.com/articles/cursor-arr-billion (2026-05-10)
• https://techcrunch.com/cursor-anysphere-funding (2026-05-10)
• https://docs.cursor.com/ (2026-05-10)