Claude Code, Cursor, GitHub Copilot and Codex compared for engineering teams

The short answer

Choose by your constraints before features. If your code lives on GitHub and you want one bill and central policies, start with GitHub Copilot. If your engineers want an AI-first editor with models from several vendors, look at Cursor. If you want Anthropic's or OpenAI's own agent across the terminal, editor and CI, compare Claude Code and Codex. Then test the two strongest candidates on changes your teams have already merged before anyone standardises.

The four tools side by side

Claude Code

Anthropic includes Claude Code in every seat of its Team plan. A Standard seat costs $20 per member per month on annual billing or $25 monthly, and a Premium seat, with five times the usage, costs $100 or $125. Beyond the allowance, if admins turn on usage credits, members draw on prepaid credits that admins can cap per organisation, group or member. On the Enterprise plan the seat covers access only, and all usage is billed at API rates.

Claude Code runs in the terminal, in VS Code and JetBrains IDEs, in a desktop app, in the cloud, and in GitHub Actions and GitLab CI. It reads CLAUDE.md files and, from version 2.1.277, AGENTS.md, which by default it uses only when there is no CLAUDE.md. Companies can run it through Amazon Bedrock, Google Cloud or Microsoft Foundry and pay per token on their own cloud bill. Anthropic states that it does not train models on code or prompts sent to Claude Code under commercial terms.

It fits teams that want to standardise on Anthropic's models and do much of their work in the terminal or in CI. Anthropic documents it only with its own Claude models.

Cursor

Cursor is an editor built around AI. Its team seats cost $30, $100 or $200 per user per month, and each tier carries a larger weekly allowance for what Cursor calls Cursor models: xAI's Grok and Cursor's own Composer. Models from Anthropic, OpenAI and Google are paid from a prepaid team balance from the first request, at list price plus a Cursor Token Rate of $0.25 per million tokens for eligible models on Teams and Enterprise. Seats created before this structure stay on legacy prices of $40 and $120.

Project rules live in .cursor/rules as .mdc files, and Cursor also reads AGENTS.md. Cloud Agents run in isolated virtual machines and open pull requests; tool execution can move to your own machines, but the agent loop stays in Cursor's cloud. Privacy Mode, under which code is not used for training, is on by default for teams.

It fits teams that want one editor with models from several vendors and accept that requests pass through Cursor's servers. Own API keys work for chat models only.

GitHub Copilot

Since 1 June 2026, Copilot bills in AI credits worth $0.01 each instead of premium requests. Copilot Business costs $19 per user per month and Copilot Enterprise $39, and each seat brings the same value in credits, pooled across the organisation. Code completions stay unlimited. Owners can set budgets per user, organisation, cost centre or enterprise, and code reviews use Actions minutes as well as credits.

Copilot reads .github/copilot-instructions.md and path-specific instruction files, and its cloud agent and CLI also read AGENTS.md or CLAUDE.md. It offers models from OpenAI, Anthropic, Google, xAI, Microsoft and Moonshot AI. Data residency in the EU or the US is available for 10% more credit consumption. GitHub's 2026 change to train on interaction data from Free, Pro and Pro+, in effect since 24 April, does not affect Business and Enterprise.

It fits organisations whose code already lives on GitHub and that want one vendor, central policies and a single bill. Copilot Enterprise requires GitHub Enterprise Cloud, and Copilot is not available for GitHub Enterprise Server.

OpenAI Codex

Codex comes with ChatGPT Business at $20 per user per month on annual billing or $25 monthly, with at least two seats. Since April 2026, usage beyond the included limits is billed in token-based workspace credits, which owners buy and limit by group. Enterprise pricing is custom. OpenAI's help centre also lists Premium Business seats at $100 or $125 with five times the usage, but that page changed on the day we checked, so confirm it.

Codex runs in the ChatGPT desktop app, a CLI, an IDE extension and the cloud, and reviews pull requests on GitHub. It reads AGENTS.md files. Its local tools can use OpenAI models through Amazon Bedrock, although cloud tasks and the admin and analytics features are not available that way. OpenAI states that it does not train on business data by default.

It fits teams that already use ChatGPT Business or Enterprise and want OpenAI's models for chat and coding in one workspace.

What matters more than the tool

The setup around an agent decides more than the logo on it: the instruction files, the tests an agent can run, the permissions it gets and the review rules for its changes. Most of that carries over between tools, because all four can read AGENTS.md. Running two tools is a legitimate choice, for example completions and reviews in one and longer agent tasks in another. Build the setup so that adding or swapping a tool later costs little.

How to decide

1. 1

   Write down your constraints

   Where your code is hosted, which cloud you must use, where data may be processed and who owns the budget.

2. 2

   Shortlist two tools

   Keep the two that meet every constraint, and drop the rest before anyone runs a demo.

3. 3

   Compare them on the same work

   Replay changes your teams have already merged, and measure acceptance, review effort and cost per merged change.

4. 4

   Set the rules before the next teams join

   Instruction files, permissions and budgets go into version control and the admin console before the rollout widens.

Designing a fair coding agent pilotWhat a rollout costs for 50 engineers