The core split is form: Claude Code is a terminal-first agent that now reaches every surface from IDE to Slack, while Cursor is a VS Code fork with agents built around the editor. Cost follows from that. Opus 5.5 is included in a $20 Claude plan, whereas in Cursor, Claude and GPT models sit outside every plan's allowance and are billed at API list price.
These two tools now overlap more than their origins suggest. Claude Code began as a command-line agent and, according to its docs, runs in the terminal, VS Code, JetBrains, a desktop app, the browser, mobile and Slack, with sessions that move between them. Cursor began as an editor and has been rearranged around an Agents Window where several local and cloud agents run side by side. The difference is the centre of gravity: Claude Code assumes you will drive it in text and wire it into scripts and CI; Cursor assumes you want to read and edit code in a graphical IDE while agents work alongside.
The cost models diverge sharply. Claude Code comes with every paid Claude plan from Pro at $20 a month, and since September 22, 2026 Opus 5.5, with a 1M-token context window, is the default even on Pro. Cursor's plans are now sold as part of an X Subscription bundle with Grok features. Plus at $8 gets only Grok models, and on Premium ($30) and above, Claude, GPT and Gemini are billed separately at API list price. A developer who mainly wants Claude or GPT in Cursor will pay well above the sticker price.
Each has a real weakness. Claude Code is locked to Claude models; you cannot drop in GPT or Gemini as you can in Cursor. Its subscription usage is shared with ordinary Claude chat, heavy agent users on Pro report moving to Max, which starts at $100 a month, and it is absent from the free Claude plan. Cursor's problem is trust in its pricing. Wikipedia's account records a July 2025 switch from 500 requests to a usage cap that ended in an apology and refunds, and a March 2026 admission that Composer 2 was built on Moonshot's Kimi model. Teams should use the soft or hard monthly spend caps its dashboard now offers.
Cursor wins on breadth around the code: Bugbot reviews pull requests on GitHub, GitLab and Bitbucket, cloud agents run in their own virtual machines and test their own changes, and its release notes show a weekly shipping pace few rivals match. Claude Code wins on depth of control, through skills, hooks, plugins, subagents and the Agent SDK. Terminal users should start with Claude Code; anyone who thinks in an editor should start with Cursor.