Coding agents compared: Claude Code, Codex, Cursor, Copilot, Devin, Jules
How the main coding agents differ in where they run, how they work with your repo, and what the plans cost, so you can shortlist two to try.
Updated 2026-11-19. "(source N)" refers to the numbered list at the end.
Two styles of coding agent
Coding agents fall into two rough groups. Some work next to you, in your terminal or editor, and ask before they change things. Others take a task, go off to their own cloud machine, and come back with a pull request. Most products now offer both, but each leans one way.
Prices below are monthly list prices in US dollars from each vendor's pricing page, checked on 19 November 2026.
Claude Code (Anthropic)
Claude Code is an agent that works through your codebase: it opens and changes files and runs commands. It runs in the terminal, in VS Code and JetBrains, in a desktop app and on the web (source 1). It connects to outside tools through MCP, reads project rules from a CLAUDE.md file or an existing AGENTS.md, and supports skills, hooks and sub-agents. It can also run on a schedule or take a bug report from Slack and return a pull request (source 1).
It is included with paid Claude plans: Pro at $20 a month, or $17 billed yearly, Max at $100, Ultra at $200, and Team seats from $20 a month billed yearly, or $25 monthly (source 2). You can also pay by the token through the API.
Codex (OpenAI)
Codex works across the ChatGPT apps, a command line tool, an editor extension and a cloud service. It is included in most ChatGPT plans (source 3).
- Free and Go ($8): Codex in the desktop app for quick or light tasks.
- Plus ($20): adds the web, CLI, editor extension, iOS and cloud integrations.
- Pro (from $100) and Business ($25 per user monthly, or $20 billed yearly) for heavier use.
- API key: use the CLI or SDK billed at API rates, handy for CI, though cloud features like GitHub code review are not available this way.
Cursor
Cursor is a code editor built around an agent. It can trace how a repo fits together, plan larger changes, fix bugs and review diffs, and it supports plugins, skills, MCP and rules (source 4). It lets you pick from many models across Anthropic, OpenAI, Google and others (source 4).
Plans run from Free, to Plus at $8 for Grok models only, Premium at $30 with cloud agents and automations, Super at $100 and Ultra at $200. Teams start at $30 per user. On the paid tiers, third-party models are billed separately at API list price (source 5).
GitHub Copilot cloud agent
Copilot's cloud agent takes a task against a repository and branch, works in a cloud environment, runs tests and linters, and can open one pull request per task. Sessions that run on GitHub Actions can change only the named repository and stop after 59 minutes (source 6). Branch protection rules can stop it from creating or updating pull requests, and GitHub advises reviewing and testing its output before production (source 6).
The free plan does not include the cloud agent. Pro costs $10, Pro+ $39 and Max $100 per user per month (source 7).
Devin (Cognition)
Devin pitches itself as an autonomous software engineer. It uses its own shell, editor and browser, and you can follow along or take over. You can reach it from a web app, a CLI, an API, or by tagging it in Slack or Teams. Its docs suggest it can handle most tasks a person could finish in about three hours (source 8).
Plans: Free, Pro at $20, Max at $200, and Teams at $80 plus $40 per developer seat. Extra usage beyond the plan is billed at API pricing (source 9).
Jules (Google)
Jules is Google's experimental coding agent. You connect GitHub, pick a repo and branch, and describe the task. Jules clones the code into a cloud machine, writes a plan for you to approve, then shows a diff and opens a pull request. It reads an AGENTS.md file if you have one (source 10).
Its page lists tiers by daily tasks: 15 tasks a day on the base tier, 100 on Pro and 300 on Ultra (source 10).
How to choose
Do not pick from a table alone. Take one real ticket from your backlog, something that takes a person half a day, and give it to your top two. Compare the pull requests: did tests pass, was the change small and readable, and how long did a person need to fix it up?
Whichever you pick, add an AGENTS.md or CLAUDE.md file with your build commands and house rules. Several of these tools read such files (source 1, source 10), so one file improves them all.
- You live in the terminal and want deep control: try Claude Code or the Codex CLI.
- You want an editor with the agent built in and a wide choice of models: try Cursor.
- Your team already runs on GitHub and wants tickets turned into pull requests: try Copilot's cloud agent.
- You want to hand off whole tickets from Slack: compare Devin with Claude Code's Slack route.
- You want a plan-first, approve-then-code flow on GitHub: try Jules.
Sources
Written by Agents Lookup in our own words. Facts come from these pages:
- 1. Claude Code Docs: Overview: code.claude.com/docs/en/overview
- 2. Claude: Pricing: claude.com/pricing
- 3. ChatGPT: Codex pricing: learn.chatgpt.com/docs/pricing
- 4. Cursor: Documentation: cursor.com/docs
- 5. Cursor: Pricing: cursor.com/pricing
- 6. GitHub Docs: About Copilot cloud agent: docs.github.com/en/copilot/concepts/agents/coding-agent/about-coding-agent
- 7. GitHub: Copilot plans: github.com/features/copilot/plans
- 8. Devin Docs: Introducing Devin: docs.devin.ai
- 9. Devin: Pricing: devin.ai/pricing
- 10. Google: Jules: jules.google