The Best AI Coding Agents in 2026: Claude Code, Codex, Copilot, Cursor and Devin

What are the best AI coding agents right now?

As of November 2026, the strongest AI coding agents are Claude Code, OpenAI Codex, GitHub’s Copilot cloud agent, Cursor’s cloud agents and Devin. A coding agent takes a task in plain English and works through your repository on its own. It reads the code, edits files, runs commands and tests, and hands back a branch or pull request for you to review.

This page covers agents only; for tools that suggest code as you type, see the best AI for coding in 2026. Prices and features come from each vendor’s pricing pages and documentation, checked on 7 and 8 November 2026. Our picks are editorial judgements based on those published details. We did not test the tools ourselves.

AI coding agents compared

Agent	Where it runs	Cheapest plan with the agent	How you hand it work
Claude Code (Anthropic)	Terminal, IDEs, desktop app, browser, Anthropic cloud VMs	Claude Pro, $20/month	Any surface, claude --cloud, @Claude in Slack
Codex (OpenAI)	ChatGPT apps, Codex CLI, IDE extension, Codex Cloud	ChatGPT Free (desktop app); Go, $8/month; cloud tasks from Plus, $20/month	ChatGPT, CLI or IDE; cloud tasks can open a pull request
Copilot cloud agent (GitHub)	GitHub-hosted cloud sandbox	Copilot Pro, $10/month	Assign an issue, or start from VS Code, the CLI, Slack or Teams
Cursor cloud agents	Isolated cloud VMs	Cursor Premium, $30/month	Cursor desktop or web, @cursor in Slack, GitHub, Bitbucket or Linear
Devin (Cognition)	Devin Cloud, or your machine via the Devin CLI	Free (light quota); Pro, $20/month, adds Devin Cloud	Web app, @Devin in Slack or Teams, CLI, API
Amp	Orbs (a machine per thread), or the CLI on your machine	Free Hobby tier (hosted orbs are pay as you go)	Web, macOS, iOS and Android apps, CLI
Google Antigravity	Desktop app, IDE and CLI (macOS, Windows, Linux)	Free tier	Desktop app that runs several agents in parallel
OpenHands	Your computer by default; Docker, VMs or your own servers	Free, open source (MIT)	Agent Canvas in your browser; scheduled or webhook automations

1. Claude Code: best overall

Anthropic describes Claude Code as “an agentic coding tool that reads your codebase, edits files, runs commands, and integrates with your development tools.” It runs in the terminal, VS Code, JetBrains IDEs, the Claude desktop app and the browser at claude.ai/code. You can also mention @Claude in Slack with a bug report and get a pull request back.

Cloud sessions each run in an isolated, Anthropic-managed virtual machine and keep going after you close your laptop. Anthropic says cloud sessions count toward your normal plan usage, with “no separate charge for the cloud VM.” You can start several in parallel, then pull a finished one back into your terminal.

• Plans: not in the free Claude plan. Pro is $20 a month ($17 billed yearly). Max is $100 for five times Pro’s usage. Ultra is $200 for ten times Pro’s weekly usage, with no five-hour limit for verified accounts.
• Models: paid plans default to Claude Opus 5.5. You can switch to Sonnet 5.5 for daily coding, Haiku 5.5 for simple jobs, or Claude Fable 5.1 for the hardest, longest-running tasks. Depending on your plan, Fable can bill to extra usage credits.

Choose Claude Code if you want one agent for both your machine and the cloud. Read our Claude Code review, or connect Claude Code to local models through Ollama.

2. OpenAI Codex: best if you already pay for ChatGPT

Codex is OpenAI’s coding agent. It works in the ChatGPT desktop, web and mobile apps, the Codex CLI, an IDE extension and Codex Cloud. Each cloud task gets its own workspace and, in OpenAI’s words, “can keep working while your computer is asleep.” You review the changed files, then commit or open a pull request.

Codex comes with every ChatGPT plan from Free to Enterprise:

• Free ($0) covers quick coding tasks and Go ($8 a month) lightweight ones, both with GPT-6 Luna in the desktop app (still rolling out).
• Plus ($20 a month) is aimed at “a few focused coding sessions each week” and adds web, CLI and IDE access, Codex Cloud and GPT-6.1 Sol. OpenAI estimates 15 to 160 local messages per five hours with GPT-6.1 Sol; cloud tasks can use more.
• Pro starts at $100 a month, with $200 and $500 tiers, and currently has no five-hour limit. Business is $25 per user a month, or $20 billed yearly.

You can also run the CLI and IDE extension with an API key at API rates, without the cloud features. Choose Codex if you already pay for ChatGPT. Torn between the two subscriptions? See Claude vs ChatGPT.

3. GitHub Copilot cloud agent: best for GitHub teams

GitHub’s Copilot coding agent is now called the Copilot cloud agent. In GitHub’s words, “Copilot can research a repository, plan changes, and implement them in the background,” by default in a GitHub-hosted cloud sandbox.

Assign an issue to Copilot and it “will start working on the task, raise a pull request, then request a review from you when it’s finished.” GitHub labels issue assignment a public preview. On 28 October 2026 GitHub began rolling out a connected cloud experience that also takes tasks from GitHub Mobile, the Copilot CLI and app, VS Code, Slack and Microsoft Teams.

• Plans: Copilot Free does not include the cloud agent. Pro is $10 a month with $15 of monthly credits, Pro+ is $39 and Max is $100.
• Billing: agent work uses AI credits (one credit is $0.01); code completions do not.
• Third-party agents: Copilot can also hand tasks to Claude and Codex agents (preview). GitHub’s docs and Pro plan card include Pro, but the comparison table on its plans page lists only Pro+ and Max, so check before you buy.

Choose Copilot if your team plans its work in GitHub and wants agent pull requests reviewed like everyone else’s.

4. Cursor cloud agents: best if you already code in Cursor

Cursor’s cloud agents, formerly Background Agents, run “in isolated VMs in the cloud.” Each clones your repository from GitHub, GitLab, Azure DevOps Services or Bitbucket Cloud, works on its own branch and pushes the changes for you to pick up. Start one from the Cursor desktop app, cursor.com/agents, the iOS app, @cursor in Slack, Linear, GitHub or Bitbucket, or the API.

• Plans: Cursor’s docs say “Free and Plus don’t include them.” Plus ($8 a month) is limited to Grok models. Premium ($30) adds “Cloud agents & automations”; Super is $100 and Ultra $200.
• Costs: cloud agents draw on your plan’s usage at the chosen model’s rate. Third-party models are billed separately at API list price, and Cursor asks you to set a spend limit first.

Choose Cursor if it is already your editor. Our Cursor review covers the editor itself.

5. Devin: best for handing off whole tickets

Cognition calls Devin “an autonomous AI software engineer that can write, run and test code.” Its site says Devin “runs in the cloud or on your machine, tests in its own browser,” and can split a big task across parallel cloud agents. Tag @Devin in Slack or Microsoft Teams, assign work in the web app, hand longer jobs from the Devin CLI to the cloud with /handoff, or use the API. It also connects to Linear, Jira and GitHub.

Plans: Free gives a “light quota to code with agents.” Pro ($20 a month) adds “Access to cloud agents (Devin Cloud)” and frontier models from several labs. Max is $200 a month; Teams is $80 a month plus $40 per developer seat. Choose Devin if you want to delegate tickets from chat.

6. Amp, Google Antigravity and OpenHands

Amp

Amp calls itself “a coding agent and development environment built for the frontier.” On the web it works in an orb, “a machine it creates for that thread,” and keeps working after you close your laptop. The CLI runs on your own machine. The free Hobby tier charges pay as you go for Amp-hosted orbs; self-hosted orbs are free, and you can use your own ChatGPT or Claude subscription or API keys. The $20-a-month Individual tier includes 45,000 minutes of hosted orb time.

Google Antigravity

Google calls Antigravity “our agentic development platform”: a desktop app that runs several agents in parallel, plus an IDE, a CLI and a Python SDK. Individuals get a free tier with “Basic weekly rate limits.” Google AI Pro ($19.99 a month in the US) raises the limits and adds a pool of AI credits; AI Ultra starts at $99.99. Our Google AI plans guide explains the tiers.

OpenHands

OpenHands is MIT-licensed. Its Agent Canvas, “the self-hosted developer control center for coding agents and automations,” runs the OpenHands agent or others such as Claude Code and Codex, on your own machine by default, through a browser interface. The README warns that without a sandbox the agent gets full access to your filesystem, so use the Docker option. To keep the model local too, see the best local LLMs for coding.

What happened to Google Jules?

Google launched Jules as “an experiment to learn how asynchronous coding agents could help you tackle your backlog.” In a changelog entry dated 29 October 2026, Google said that, to unify its agentic coding efforts, “Jules will wind down on December 31, 2026.” Sessions could be finished until 11 November 2026; after that, Jules is read-only until the end of the year. You can export your Jules data to a local folder; code and pull requests stay in GitHub. Google says lessons from Jules are shaping its agent-first platform, Antigravity.

How to choose an AI coding agent

• Strongest all-rounder: Claude Code on Claude Pro.
• Already paying for ChatGPT: try Codex before buying anything else.
• Team lives in GitHub Issues: Copilot cloud agent, from $10 a month.
• Delegating from chat: Devin, Claude Code, Cursor, Copilot and Codex (from ChatGPT Plus) all connect to Slack.
• Nothing up front: Codex on ChatGPT Free, Antigravity, Devin Free, Amp Hobby or self-hosted OpenHands.
• Not a coder? Our best AI agents guide covers agents for everyday tasks.

Before you let an agent loose on your code

• Prefer sandboxes. The cloud agents from Anthropic, OpenAI, GitHub and Cursor work in separate cloud environments. Run local or self-hosted agents in a container.
• Review every pull request. Agents make confident mistakes, so run your tests before merging.
• Guard secrets and spending. Anthropic says GitHub credentials for its hosted cloud sessions never enter the session’s VM; check how your agent handles them. Parallel tasks use up credits faster, so set spend caps.

FAQ

What is an AI coding agent?

Software that takes a plain-English task and completes it across your codebase with little supervision: it edits files, runs commands and tests, fixes its own errors and hands back a branch or pull request. A coding assistant mainly suggests code as you type.

Which AI coding agent is free?

Free options: Codex in ChatGPT Free (desktop app), Google Antigravity’s free tier, Devin Free’s light quota, Amp’s Hobby tier (hosted orbs are pay as you go) and open-source OpenHands. Claude Code and Copilot’s cloud agent need a paid plan.

Is Claude Code or Codex better?

Claude Pro and ChatGPT Plus both cost $20 a month, and on those plans both agents work locally and in the cloud. Claude Code defaults to Opus 5.5; Codex is also included in ChatGPT Free and Go. If you already pay for one subscription, start with the agent it includes.

Is Google Jules shutting down?

Yes. Google’s changelog says Jules winds down on 31 December 2026 and is read-only after 11 November 2026 until then. You can export your Jules data to a local folder.

Does my code need to be on GitHub?

It depends. Copilot’s cloud agent and Claude Code’s Anthropic-hosted cloud sessions need GitHub to clone repositories and open pull requests. Cursor’s cloud agents also clone from GitLab, Azure DevOps Services and Bitbucket Cloud. Local CLIs such as Claude Code, Codex and Amp work with files on your computer.