# Claude Code, Codex or Copilot: how do coding agents differ?

Source: https://codelifter.com/forum/posts/claude-code-codex-or-copilot-how-do-coding-agents-differ/
Site: CodeLifter Forum, category MCP and tools
Published: 2026-10-30. Updated: 2026-11-04. Replies: 2.

## Question by CodeLifter editors (staff), 2026-10-30

Claude Code, Codex and GitHub Copilot are all coding agents that read a repository, edit files and run commands for you. They differ in maker, licence, the models they run, where the work happens and how they fence it in. No neutral, published test settles which is better, so choose on those differences and try each on your own code.

## What is a coding agent?

A coding agent is an assistant that does more than suggest text: it reads files, changes them, runs commands such as tests, and repeats until a task is done. Anthropic's [Claude Code overview](https://code.claude.com/docs/en/overview) describes it as a tool that "reads your codebase, edits files, runs commands, and integrates with your development tools". OpenAI's [Codex CLI page](https://learn.chatgpt.com/docs/codex/cli) says Codex can "inspect files, make edits, and run the tools already installed on your machine". GitHub's [Copilot CLI page](https://docs.github.com/en/copilot/concepts/agents/about-copilot-cli) calls its command-line client "a powerful AI agent" that can make changes and create a pull request.

## How do they compare?

Everything below comes from each vendor's own documentation, read on 15 and 16 October 2026. These products change monthly, so check the linked pages before you rely on a row.

| | Claude Code | Codex | GitHub Copilot |
| --- | --- | --- | --- |
| Maker | Anthropic | OpenAI | GitHub |
| Licence | Proprietary ("All rights reserved" in its LICENSE.md) | Codex CLI is Apache-2.0 | VS Code chat client is MIT; the service needs a Copilot plan |
| Where it runs | Terminal, IDE extensions, desktop app, web | CLI, IDE extension, ChatGPT desktop app, Codex cloud | IDE agent mode, Copilot CLI (`copilot`), cloud agent on GitHub.com |
| Models | Claude models via Anthropic, Amazon Bedrock, Google Cloud or Microsoft Foundry | OpenAI models; built-in `ollama` and `lmstudio` providers | Models from Anthropic, OpenAI, Google and xAI |
| Sign-in | Claude subscription or Anthropic Console account | ChatGPT sign-in or an API key | GitHub account with a Copilot plan, including Copilot Free |
| Rules file | `CLAUDE.md`, or `AGENTS.md` when there is no `CLAUDE.md` | `AGENTS.md` (or `AGENTS.override.md`) | `.github/copilot-instructions.md` and `AGENTS.md` |
| Skills folder | `.claude/skills/` | `.agents/skills/` | `.github/skills/`, `.claude/skills/` or `.agents/skills/` |
| MCP servers | Yes | Yes | Yes |
| Default guard | Permission modes; Manual mode asks before most edits, commands and network calls | Sandbox applied by default to every command it spawns | CLI needs approval options before it may change or run files unasked; cloud agent works in an ephemeral cloud environment |

The table leaves out two things on purpose: prices, which change and differ by plan, and speed or quality, which no vendor page measures against the others.

## Where do they really differ?

**Where the work happens.** Copilot's cloud agent, per [GitHub's docs](https://docs.github.com/en/copilot/concepts/agents/cloud-agent/about-cloud-agent), "can edit files and run tests and linters in an ephemeral cloud development environment" and handles "branch creation, commit messages, and pushing changes". Codex also has a cloud mode, which its [authentication page](https://learn.chatgpt.com/docs/auth) says "requires signing in with ChatGPT". Claude Code runs locally by default and offers web sessions as well. If your code cannot leave your machine, the local modes are what matter.

**How each one is fenced in.** Codex's [sandbox page](https://learn.chatgpt.com/docs/sandboxing) says local commands "run inside a constrained environment instead of running with full access by default", using Seatbelt on macOS and bubblewrap on Linux and WSL2. Claude Code instead relies on [permission modes](https://code.claude.com/docs/en/permission-modes): in Manual mode it "stops and asks you before most actions that edit files, run shell commands, or reach the network", and from v2.1.283 the interactive terminal starts in auto mode, where a second model reviews actions instead of you. Check which mode you are in before a long task.

**Which models you can use.** Copilot lets you pick models from several companies. Codex accepts custom providers in `config.toml`. Claude Code runs Claude models, through Anthropic or a cloud provider. If your employer has approved one model vendor, that may decide it for you; the [employer policy thread](/forum/posts/do-employers-allow-ai-coding-assistants-and-what-do-their-policies-say/) covers that side.

## So is Codex better than Claude Code?

No published source answers that head to head. The vendor pages describe features, not head-to-head results. The scores on the CodeLifter [models page](https://codelifter.com/models/) rate model answers on a chat leaderboard; they say nothing about how an agent edits a real repository. Treat any ranking without a published method as an opinion.

## How do you choose for your own project?

1. List where your code lives. If issues and pull requests are on GitHub and you want to assign work from there, Copilot's cloud agent fits that flow.
2. Check which model vendors your school, employer or client allows.
3. Decide how much the agent may do without asking, and pick the tool whose default guard matches.
4. Write one `AGENTS.md`. All three can read it, though Claude Code reads it only when there is no `CLAUDE.md`. The [AGENTS.md thread](/forum/posts/what-should-go-in-an-agentsmd-or-coding-rules-file/) covers what to put in it.
5. Give each tool the same small task on its own branch, then compare the diffs and run your tests.
6. Review the result as you would any code you did not write, line by line, before you merge it.

The full side-by-side of twelve agents, with licence and rules-file sources, is in the CodeLifter [coding agents directory](https://codelifter.com/agents/).

## Short version

- All three read code, edit files, run commands and connect to MCP servers.
- Codex CLI is Apache-2.0; Claude Code is proprietary; Copilot is a paid or free GitHub service.
- They differ most in where work runs, which models they use and how they are fenced in.
- No neutral test says which is better; try the same task in each on your own repository.
- One `AGENTS.md` file gives all three the same rules.

## Reply 1 by CodeLifter editors (staff), 2026-10-30

**Can you use Claude models in Copilot or Codex, or OpenAI models in Claude Code?** In Copilot, yes: GitHub's [supported models table](https://docs.github.com/en/copilot/reference/ai-models/supported-models) lists models from Anthropic, OpenAI, Google and xAI, and which ones you see depends on your plan. On Copilot Free and Copilot Student, the [plans page](https://docs.github.com/en/copilot/get-started/plans) (read 16 October 2026) says models are available "through auto model selection only", so you cannot pick one. Codex is built around OpenAI models, but its [advanced configuration page](https://learn.chatgpt.com/docs/config-file/config-advanced) lets you define other providers under `[model_providers]` with a `base_url` and an `env_key`; its own examples are a proxy, a local Ollama server and Mistral. Whether a given model works well depends on that provider's API and the model's tool calling, which the page does not promise. Claude Code's [model configuration page](https://code.claude.com/docs/en/model-config) covers Claude models through the Anthropic API, Amazon Bedrock, Google Cloud's Agent Platform and Microsoft Foundry. Pointing Claude Code at anything else means an LLM gateway set with `ANTHROPIC_BASE_URL`, and Anthropic warns that a gateway which does not forward new features "breaks the corresponding features".

## Reply 2 by CodeLifter editors (staff), 2026-11-04

**Which coding agents are open source?** Of the three in this thread, only the Codex CLI is: its [GitHub repository](https://github.com/openai/codex) is licensed Apache-2.0. Claude Code's [LICENSE.md](https://github.com/anthropics/claude-code/blob/main/LICENSE.md) reads "All rights reserved." GitHub publishes the Copilot chat client for VS Code under MIT in [microsoft/vscode-copilot-chat](https://github.com/microsoft/vscode-copilot-chat), but that repository notes an active Copilot subscription is required, so the service itself is not open. Outside the big three, the CodeLifter agents directory checked these licences on 15 October 2026: [Gemini CLI](https://github.com/google-gemini/gemini-cli) is Apache 2.0, [Aider](https://github.com/Aider-AI/aider) and [Cline](https://github.com/cline/cline) are Apache-2.0, and Zed is GPL-3.0-or-later with Apache-2.0 parts. Continue is Apache-2.0 but its repository now says it is "no longer actively maintained and is read-only". One caution: an open source client can still send your code to a hosted model. The licence tells you what you may do with the agent's code, not where your prompts go. For an agent that runs end to end on your own machine, see the local model thread.

---
Cite as: CodeLifter Forum, "Claude Code, Codex or Copilot: how do coding agents differ?", https://codelifter.com/forum/posts/claude-code-codex-or-copilot-how-do-coding-agents-differ/, accessed 2026-11-12.