Two different kinds of product
ChatGPT is OpenAI's general assistant, and Codex, its coding agent, is included in the same plan. Cursor describes itself as a coding agent for building software; its general chatbot, Grok Chat, comes from the X Subscription bundle, not from the editor. xAI says Premium bundles SuperGrok, Cursor Pro and X Premium+. So on ChatGPT, code is one part of what you pay for; on Cursor, the editor and its agents are, with Grok and X added.
Where you write code
Codex reaches your code through an IDE extension that signs in with ChatGPT and runs in VS Code, VS Code Insiders, Cursor and Windsurf, with Xcode and JetBrains IDEs through their own integrations. Its usage is shared with ChatGPT Work. OpenAI documents it as an agent beside your code: chat with editor context, review diffs, delegate to the cloud. Completions as you type are not documented for Codex, and no page we checked says they are absent, so that is not verified.
Cursor is the editor itself, with Tab autocomplete for multi-line and cross-file suggestions and next-edit jumps. Cursor Agent edits files, runs terminal commands and searches code. Cursor also documents a command line tool, web agents at cursor.com/agents, an iOS and iPadOS app, its agent in JetBrains IDEs 2025.1 or later through ACP, and Xcode 26.3 or later through an MCP bridge.
Cloud agents and pull request review
On Plus, Codex Cloud runs coding tasks on your repos in OpenAI-managed cloud environments and keeps going while your computer sleeps; when it is done you can commit or open a pull request. OpenAI says this is subject to rollout for eligible Plus accounts, and that usage counts toward plan limits. Codex code review works on GitHub pull requests: comment @codex review or turn on automatic reviews, and it posts a review flagging P0 and P1 issues, with an optional Security Review in research preview.
On Premium, Cursor's cloud agents run in cloud machines on GitHub, GitLab, Bitbucket or Azure DevOps repos, work on a branch and open pull requests, and can be started from the web, Slack, GitHub comments or the iOS app. Bugbot reviews GitHub and GitLab pull requests automatically on each update. Which code hosts Codex Cloud supports is not stated in the facts we collected. Whether either agent's changes work is a question for testing: run our code repair, code spec and code refactor kits on both.
Models, context and limits
ChatGPT Plus lists OpenAI models only: GPT-6.1 Sol, GPT-6 Astra, GPT-6 Sol, GPT-6 Luna, the GPT-5.6 family and legacy models, with limited GPT-6 Astra use in Codex and optional credits. OpenAI documents 256K tokens of context for reasoning models and 54K for Instant, and describes Plus limits as expanded, without figures.
Cursor Premium includes Grok 4.7 and Composer 2.5 in a pool 4 times the size of Plus, refreshed weekly and shared with Grok. Models from Anthropic, OpenAI, Google and Meta, among them GPT-5.6 Sol, Claude Opus 5.5 and Gemini 3.1 Pro, are billed on top at API list price, with spending caps. Grok 4.7 has a 256K window and 500K long context; Claude Opus 5.5 goes up to 1 million tokens, paid per use.
Everything that is not code
ChatGPT Plus documents image generation, voice with video, expanded deep research, data analysis with interactive tables and charts, file uploads, projects with scheduled tasks, memory, extensions for Excel, Word, PowerPoint and Google Sheets, and the ChatGPT Work agent on desktop, web and mobile. Video generation is not offered on any ChatGPT plan.
In Cursor itself, chat uploads are documented for images only, and research reports and spreadsheet analysis are not documented. The general features come from Grok in the bundle: Grok Chat, Grok Imagine for images, Grok Voice and, from Premium, Grok video generation, all drawing on the same pool as your coding.
Privacy, ads and export
ChatGPT uses your chats for training by default; turn off "Improve the model for everyone" under Settings, Data controls. You can request a copy of your data, and Plus has no ads. Cursor's Privacy Mode keeps code data out of training by Cursor and model providers; with it off, Cursor may train on codebase data and prompts. Its default is not documented on the pages we checked, and neither is a data export. No ads are documented inside Cursor; on X, Premium is ad-free on "most surfaces".
Free tiers and other plans
ChatGPT Free and Go ($8 a month) list Codex as limited, but OpenAI documents Codex Cloud as not included on either, does not state whether pull request review is, and conflicts on the editor extension. Both may show ads. The Pro tiers ($100, $200 and $500 a month) document expanded Codex with longer sessions and, currently, no five-hour limit.
Cursor Free has limited agent and chat requests and no cloud agents, Bugbot or third-party models. Plus, at an introductory $8 a month, has Grok models only and still no cloud agents or Bugbot. Cursor does not state a Tab limit for Free or Plus, so that is not verified.
How this page was made
The What the tech desk wrote this page with AI assistance, from the facts in our data release; our methodology explains how they are collected. Every plan is on the ChatGPT and Cursor pages. If you are also weighing Copilot, see Cursor vs GitHub Copilot.