Cursor

  • Cursor — AI-Native VS Code Fork With Tab Completion, Composer Multi-File Edits, and Cloud Background Agents (Anysphere, In-Editor Driver-First, Not a Terminal/Cloud-Only Agent Like Claude Code/Devin)

     

    1. One-line positioning

    Cursor (San Francisco, 2022; Anysphere — Michael Truell, Sualeh Asif, Arvid Lunnemark, Aman Sanger; valued ~$60B mid-2026) = the AI-native VS Code fork that made agentic coding mainstream — Tab autocomplete (Sub-100ms, multi-line) + Composer (multi-file diff plan) + Agent mode (explore→plan→run tests→fix lint) + Cloud/Background agents (up to 8 parallel in 2.0) + Bugbot PR review + MCP/skills/hooks + bring-your-own-key, routing across GPT-5.4 / Claude Opus 4.6 / Gemini 3 Pro / Grok Code / Composer 2.5. Not GitHub Copilot (plugin, cheaper, shallower), not Claude Code (terminal-first, no IDE), not Windsurf (SWE-1.5 only, same price), not JetBrains AI (different editor family). Fits ai-productivity-data; pipeline_stage none (spans code-generation + refactor + agent-execution + pr-review). Table-first, Claude Code skeleton. The 2026 trap: June-2025 credit-meter reprice — $20 Pro includes ~$20 of frontier-model API usage, Auto mode unlimited but manual frontier picks burn credits, heavy agent users report $40–100/mo effective spend.

    2. Core features

    Layer Capability Notes
    Editor VS Code fork, same keymaps/extensions/themes, macOS/Win/Linux No JetBrains/Vim/Emacs core; web only, no mobile IDE
    Tab Supermaven-based multi-line autocomplete, auto-import, next-edit prediction Sub-100ms, best-in-class per G2/Capterra
    Chat Cmd-K inline + sidebar chat, full-codebase @-references, image paste Model-aware (Claude/GPT/Gemini/Grok/Composer)
    Composer Multi-file edit planner, unified diff view, accept/reject per hunk 2.0 supports up to 8 parallel agents
    Agent Autonomous: explore repo → plan → edit → run tests → read terminal → fix lint → iterate Background + Cloud agents (surcharge 20% on cloud)
    Models Auto Cost / Auto Balance / Auto Intelligence + manual pick; BYOK supported Auto Cost $1.25/M in, $6/M out, $0.25/M cache read; third-party models + Cursor Token Rate $0.25/M
    Ecosystem MCP servers, skills, hooks, Bugbot (GitHub PR review), 2,000+ integrations Bugbot $40/user Pro, $40/user Teams, separate from editor seat
    Surface 🌐 Web docs · 🖥️ Desktop (mac/win/linux) · 📱 iOS app (monitor/steer agents only, no editing) · 🔌 REST/API (enterprise) · 💻 no native CLI (Cursor CLI beta) —

    3. Pricing (2026-08, USD/mo)

    Tier Price Seats / credits Unlocks
    Hobby $0 1 user, 2,000 Tab completions, 50 slow premium requests/mo Editor, basic chat, no Cloud agents, legacy models only
    Pro $20 ($16 ann) 1 user, ~$20 frontier-model API pool, unlimited Tab, unlimited Auto All models manual pick (burns pool), Composer, Agent, MCP, Cloud agents (overage)
    Pro+ $60 ($48 ann) 1 user, ~3x pool (~$60) Same features, 3x usage
    Ultra $200 ($160 ann) 1 user, ~20x pool (~$400) Priority models, experimental access
    Teams $40/user ($32 ann) Standard · $120/user Premium Standard seat +$20 pool; Premium 5x Standard agent limits Shared rules/chats, SSO/SAML/OIDC, RBAC, usage analytics, privacy-mode enforce, Bugbot team
    Enterprise Custom Pooled usage, SCIM, audit logs, AI code tracking API, invoice/PO SSO, priority support, custom model controls
    Bugbot (add-on) Pro $40/user · Teams $40/user · Enterprise custom — PR review on up to 200 PRs (Pro) / all PRs (Teams)

    pricing_type = freemium; starting_price = $0; free_tier = "Hobby: 1 seat, 2000 Tab completions/mo, 50 slow premium requests/mo, no Cloud agents, legacy models (GPT-3.5/Claude 3 class)"; pricing_note = "june-2025-credit-meter-reprice; pro-20-includes-20-frontier-api-pool; auto-mode-unlimited-but-manual-pick-burns-pool; heavy-agent-users-40-100-effective; cloud-agents-20pct-surcharge; cursor-token-rate-0.25-per-M-on-third-party; teams-standard-40-premium-120; bugbot-40-separate-seat; annual-20pct-off; no-reseller"

    4. Access Type (v2)

    access_type: desktop-app, api

    access_display: 🖥️ macOS/Win/Linux desktop · 📱 iOS monitor-only · 🔌 MCP / REST API (Enterprise) · 💻 Cursor CLI (beta)

    5. Reviews (split — G2/Trustpilot divide)

    Platform Score Volume Signal
    G2 4.5–4.8/5​ ~255–3,000 Tab autocomplete, Composer, codebase context, multi-model switch
    Capterra 4.7–4.8/5​ ~400 SMB dev usability, onboarding from VS Code
    Trustpilot 1.6–1.9/5​ ~285 June-2025 billing shock, overage disputes, refund denials, support ghosting
    Product Hunt 4.7/5 — Early-adopter enthusiasm
    Aggregate ~4.6/5​ (G2+Capterra) — Trustpilot gap = billing/credit-meter pattern (same shape as Ahrefs/ClickUp/Character.AI)

    Recurring praise: Tab multi-line prediction, Composer cross-file refactors, Agent runs tests/fixes lint, BYOK control, VS Code muscle-memory zero-cost migration.

    Recurring gripes: credit meter opaque (Pro $20 evaporates in heavy Opus use), silent-revert bug early 2026 (patched), large-monorepo index lag, memory/CPU spike on long agent sessions, Trustpilot billing/auto-renew fights, 5-day refund window on annual.

    6. Best for / Not for

    Best for Not for
    Individual devs in VS Code doing multi-file refactors daily Budget-predictable teams wanting flat $10 seat (Copilot cheaper)
    Agent-mode power users (Pro+ $60 covers most) Terminal-first engineers who live in Claude Code / Codex CLI
    Startups shipping features with parallel Background agents JetBrains/Vim/Emacs-only shops (no native port)
    Teams wanting SSO + shared rules + Bugbot on PRs (Teams $40) Regulated/on-prem/air-gapped (Tabnine self-hosted beats it)
    Agencies swapping Claude/GPT/Gemini per task Solo bloggers / non-devs (zero value)
    Proprietary-code cos that enable Privacy Mode (Business+) Users who manually pick Opus for every keystroke (bill shock)

    7. Competitors (table)

    Tool Lane Starts vs Cursor
    GitHub Copilot Plugin for any IDE $10/mo ($19 Business) Cheaper, shallower repo context, no Composer/Agent parity, no multi-model manual pick
    Claude Code Terminal agent $20 Pro / $100 Max 5x / $200 Max 20x Better autonomous terminal execution, no IDE/Tab/diff UI, pairs with Cursor
    Windsurf AI-native IDE (SWE-1.5) $20/mo Pro · $40 Teams Same price tiers, proprietary model only, Cursor has broader model switch + bigger mindshare
    Tabnine Self-hosted enterprise coding $39/user (teams only) On-prem/air-gapped/IP indemnity, weaker agentic IDE, no Composer
    Trae (ByteDance) AI IDE (CN/INTL) CN free · INTL $3–100 Cheaper/free CN, less mature agent, smaller Western enterprise trust
    Zed Rust-native collaborative IDE $20/mo (some free) Faster idle editor, thinner agent/PR-review ecosystem
    Codex CLI / OpenAI Terminal coding API-priced No IDE, pairs as backend under Cursor

     

Do Not Sell or Share My Personal Information Cookie Settings