Cursor

Cursor is a VS Code fork with AI baked into the editor — not as a sidebar, but as a first-class layer that drives completions, refactors, codebase chat, and an agent mode that can run multi-file edits autonomously. It's the most-used AI coding tool by paying customer count in 2026, with 360,000+ paying users and $2B in annualised revenue. The user base alone is a real signal — when a category of tool gets to that scale, it's because the product works for the bulk of the addressable market.

The headline differentiator is multi-model selection. Claude Sonnet 4.6, GPT-5 family, Gemini 2.5 Pro and Flash, plus a handful of cheaper alternatives are all available — and you can switch between them mid-session for the same task. Try Claude for a complex refactor; if it doesn't get it right, switch to GPT-5 and see if it handles the architecture differently. No other dominant coding tool in 2026 supports this without environment-switching. Claude Code only runs Anthropic; OpenAI Codex only runs GPT; the open-source CLI tools (Aider, Cline) support multi-model but lack the IDE integration.

Pricing is the part most builders need to understand before subscribing. The Hobby tier is free with 2,000 completions and 50 slow requests per month — usable for casual experimentation, not for serious daily work. Pro at $20/month is the tier most builders settle on; the headline is unlimited completions plus a credit pool for agent and chat usage. Pro+ at $60/month and Ultra at $200/month give you bigger pools. Teams is $40/user/month with admin features and shared billing. The complication: in June 2025, Cursor replaced fixed fast-request quotas with a $20 credit pool, which effectively cut monthly fast requests by roughly 55% for most users. The change wasn't announced clearly; existing Pro subscribers found their monthly capacity quietly shrunk. Trust dented — and older reviews predating the change overstate what $20 buys today.

What it's actually like to use, day to day: Tab completion is genuinely best-in-class. The AI suggests not just the next line but the next several edits across multiple files, and you accept-or-reject with one keystroke. `Cmd+K` for in-line edits is the fastest way to make a targeted change that exists in any coding tool we've tested. The chat sidebar is competent — codebase-aware questions get accurate answers because Cursor indexes your repo. Agent mode is the part where opinions split: for short tasks (rename a function and update call sites, add a route, write a test), agent mode works well. For longer agentic loops on real codebases, Claude Code is more reliable in our testing — Cursor's agent mode occasionally makes large unreviewable multi-file edits that are hard to roll back. Pair with branch-and-commit-first discipline.

Which model to pick on Cursor depends on the task. Claude Sonnet 4.6 is our default for everything code-shaped — best reasoning per dollar in the lineup. GPT-5 is the right swap when Claude struggles on architectural decisions or unfamiliar frameworks. Gemini 2.5 Pro is the right swap for very long-context tasks (multi-file refactors across large codebases) where Cursor's context window matters. The cheaper models (Claude Haiku, GPT-5 Mini, Gemini Flash) are competent for completions but we wouldn't run agent mode on them for production work. The model selector is in the bottom-right of the chat panel — most builders set it once and forget it; the value compounds when you actually swap based on task shape.

Where Cursor is the wrong pick: if you're on JetBrains, Vim, or any non-VS Code editor, Cursor requires you to switch IDEs — usually not worth it for what you gain. If your workflow is CLI-first or terminal-native, Aider and Claude Code give you the same model access without the IDE. If you're a high-volume user where direct Claude API costs would be cheaper than Cursor Pro's effective rate, switching to Cline (BYOK, open-source, same Claude API) often wins on economics. And if SaaS pricing volatility is something you can't live with, the June 2025 pricing change is a real warning sign that should factor into the choice.