The short answer
The best GitHub Copilot alternative in November 2026 depends on why you are leaving, but for most developers it is Anthropic's Claude Code, because Claude Opus 5.5 and Sonnet 5.5 come inside a Claude Pro, Max or Ultra plan instead of a token-priced credit balance. Pick Cursor for an agent-first editor with Grok 4.7 included, OpenAI Codex if you already pay for ChatGPT, Devin Desktop or Zed for a different editor, and Replit if you are not a developer.
Why people look for GitHub Copilot alternatives
GitHub Copilot is still the default AI coding tool in many companies, and for good reasons: it works in almost every IDE and it lives where the pull requests are. But 2026 brought changes that make some people reconsider.
- Usage-based billing. Since 1 June 2026 every plan bills chats and agent runs in GitHub AI Credits, based on tokens at each model's API rate. Inline completions and Next Edit suggestions are free of credits, but a long agent session on an expensive model can use a month's allowance quickly, and there is no automatic fallback to a cheaper model when credits run out.
- Training on individual plans. Since 24 April 2026, interaction data from Copilot Free, Pro, Pro+ and Max can be used to train AI models unless you opt out. Business and Enterprise are excluded.
- Code review costs twice. Since 1 June 2026, Copilot code review uses GitHub Actions minutes as well as AI credits.
- GitHub only. The cloud agent works only on repositories hosted on GitHub, one branch and one pull request per task, with Actions-based sessions capped at 59 minutes.
Before you leave, note what Copilot does well that others do not: it works in Visual Studio, JetBrains IDEs, Xcode and Eclipse as well as VS Code, Copilot Fusion (generally available since 28 October 2026) routes work across models with no separate charge beyond the models it uses, and Copilot Free still offers 2,000 completions a month.
The six alternatives at a glance
| Tool | Best for | Free plan | Main catch |
|---|---|---|---|
| Claude Code | Claude models inside a flat subscription | No (Claude Pro or above) | Weekly usage limits |
| Cursor | Agent-first editing, parallel agents | Yes, limited, no third-party models | Claude and GPT cost extra, Premium and above |
| OpenAI Codex | ChatGPT subscribers | Yes, desktop app only | CLI, IDE and cloud start at Plus |
| Devin Desktop | A full agent editor from Cognition | Yes, light agent quota | Renamed from Windsurf in June 2026 |
| Zed | A fast editor with your own keys | Yes, with your own keys or agents | No hosted models on free |
| Replit | Building a hosted app without coding | Yes, Starter | Free apps go down after 30 days |
1. Claude Code: the pick for predictable access to Claude
Best for: developers who mostly use Claude models in Copilot and want them without a token-priced balance.
Why: Anthropic's Claude Code is a coding agent that reads your codebase, edits files, runs commands and works with git. It runs in the terminal, in VS Code and JetBrains, in the Claude desktop app, on the web at claude.ai/code and from Slack, and works with GitHub Actions and GitLab CI/CD. Claude Opus 5.5, Sonnet 5.5 and Haiku 5.5 are included in your plan's usage limits. CLAUDE.md, skills, hooks and subagents let you shape how it works, and cloud scheduled tasks keep running when your computer is off. Code Review became generally available for Team and Enterprise on 15 October 2026.
The catch: every plan has usage limits, and on Free, Pro, Max and Ultra your coding sessions can be used for training unless you switch off 'Help improve our AI models' in Settings, then Privacy.
Free plan: none for Claude Code; it starts with Claude Pro. Claude Free still helps with snippets in chat.
2. Cursor: the pick for agent-first editing
Best for: developers who want an editor built around agents rather than an assistant added to one.
Why: Cursor's Agents Window runs several agents in parallel, locally, in git worktrees, in the cloud or over SSH. /best-of-n compares models on the same task, Plan Mode writes an editable plan first, cloud agents can hold a long-running /goal, and Cursor Projects (beta) coordinates many agents on a migration. Paid plans include Grok 4.7, Grok 4.6, Grok 4.5 and Composer 2.5 in a weekly pool. With Privacy Mode on, Cursor says your code is not used for training by Cursor or its providers.
The catch: since SpaceX bought Cursor (14 August 2026) and moved it into The X Subscription (12 October 2026), Claude, GPT and Gemini are only on Premium and above, at API list price on top of the plan. If you used Copilot for its model choice, Cursor is a step back unless Grok suits you.
Free plan: limited Agent and Chat requests, no third-party models. See Cursor vs GitHub Copilot.
3. OpenAI Codex: the pick if you already pay for ChatGPT
Best for: people on ChatGPT Plus, Pro or Business who mostly used GPT models in Copilot.
Why: Codex is included in ChatGPT plans and runs in the ChatGPT desktop and mobile apps, on the web, in the Codex CLI, the Codex IDE extension and Codex Cloud. From Plus you get the web, CLI, IDE extension and iOS, plus automatic code review on GitHub when you tag @Codex on a pull request, and Slack. GPT-6.1 Sol and GPT-6.1 Terra are available on Plus and above.
The catch: the Free and Go plans only get the desktop app with GPT-6 Luna. Code review through GitHub counts against its own limits, and local reviews count against your general usage.
Free plan: yes, for quick coding tasks in the desktop app, subject to rollout.
4. Devin Desktop (formerly Windsurf): a full agent editor
Best for: developers who want to swap Copilot in VS Code for a complete agent editor with cloud agents.
Why: Cognition renamed Windsurf to Devin Desktop on 2 June 2026. It is built on the Windsurf IDE, with an Agent Command Center, a Kanban view for local and cloud agents, Spaces for shared context and Devin Review for diffs. The new Devin Local agent replaced Cascade. Pro adds frontier models from OpenAI, Anthropic and Google, cloud agents and extra usage at API pricing.
The catch: the product changed name and default agent this year, and its pricing page does not state its training or retention policy, so check before using it on company code.
Free plan: yes: a light agent quota, limited models, unlimited Tab completions and inline edits.
5. Zed: the pick for control over models and cost
Best for: developers who want a fast editor and to pay their model provider directly.
Why: Zed's free Personal plan allows unlimited use with your own API keys or external agents such as Claude Agent and Codex CLI. Pro adds Zed-hosted models and unlimited edit predictions; Business adds admin controls for model policies and data sharing. Zed says it does not keep your prompts or code context by default.
The catch: Zed is not VS Code, so your extensions and keyboard habits need rebuilding, and on the free plan you still pay for the model through your own keys.
Free plan: yes, Personal is free forever.
6. Replit: the pick for non-developers
Best for: people who used Copilot to stumble through code they do not really want to write.
Why: Replit Agent 4 builds and hosts complete apps in the browser from plain language, with databases, storage and publishing built in. Models include Claude Opus 5.5, Sonnet 5.5 and GPT-6.1 Sol depending on mode and plan.
The catch: credit-based Power and Max modes make costs hard to predict, and the training opt-out is only documented for Pro and Enterprise.
Free plan: Starter, with daily Agent credits and one published app that goes down after 30 days.
Which one should you choose?
- You are leaving because of unpredictable credits: Claude Code on Pro or Max, because usage comes as plan limits rather than a token balance.
- You are leaving because of training on individual plans: either opt out in Copilot (it takes a minute) or move to Copilot Business; if you still want to switch, Zed with your own keys gives the most control.
- You want a more powerful agent editor: Cursor or Devin Desktop.
- You already pay for ChatGPT: OpenAI Codex.
- You do not code: Replit.
Do not leave Copilot just to get a model: Copilot's paid plans include Claude Opus 5.5, GPT-6.1 Sol and Gemini 4 Argon, depending on plan and admin policy. Leave if the billing model, the editor or the workflow does not fit.
How to switch from GitHub Copilot without losing your work
- Opt out of training first. Click your profile picture on GitHub, open Copilot settings and set 'Allow GitHub to use my data for AI model training' to Disabled, even if you plan to cancel.
- Finish open agent work. Merge or close pull requests opened by the Copilot cloud agent and stop any Copilot automations, so nothing keeps spending credits.
- Move your instructions. Copy any repository instructions you wrote for Copilot into a plain AGENTS.md at the repository root. Claude Code can read AGENTS.md in place of CLAUDE.md, and Cursor reads AGENTS.md too.
- Re-add MCP servers. Copilot supports MCP in agent mode; list your servers and add them to the new tool; Claude Code and Cursor both support MCP servers.
- Replace code review. If your team relied on Copilot code review, set up the replacement (Claude Code in GitHub Actions, Cursor's Bugbot or Codex review with @Codex) before you switch Copilot off.
- Disable the extension, do not just ignore it. Two AI assistants competing for the same keystrokes in one editor is confusing; switch off the Copilot extension during your trial.
- Cancel at the end of the billing cycle, after a week of giving the same tasks to both tools.
Read our GitHub Copilot guide, Cursor alternatives, the best AI for coding, or line up two tools in the comparison tool.
Prompts to copy
Estimate whether you need a flat plan
Here is my last month of Copilot usage: [paste the AI credits usage summary or describe tasks]. Group it into completions, chat and agent runs, tell me which tasks used the most credits and whether a flat subscription with weekly limits would suit this pattern better. Ask me for anything you need.
You get: A clear view of where your credits went
Write a portable instructions file
Turn these Copilot repository instructions into an AGENTS.md file for the repository root: [paste instructions]. Keep every convention, add the commands to install, test and lint, and list what an agent must never touch. Keep it under 70 lines.
You get: One instructions file most coding agents can read