AI Coding Assistants in 2026: Copilot, Cursor, Claude Code, Codex and Antigravity Compared

Every AI coding tool now promises an "agent" that plans, edits files and opens pull requests. The marketing pages have started to look the same, which makes picking one harder than it was two years ago. The differences that matter are easy to check once you know where to look: how each tool charges you, where it runs, which models you can use, and what happens to your code.

This guide compares five AI coding assistants that working developers are likely to shortlist. Every price below comes from the vendor's own pricing page as of November 2026. Plans and prices change often, so treat the numbers as a snapshot and confirm them before you pay.

I mostly use ChatGPT for coding help, because it explains things clearly and helps me debug faster.

Start with how you work

Before comparing tools, write down three things:

1. Where you spend your day: VS Code, a JetBrains IDE, the terminal, or the GitHub web UI.
2. What you want help with: inline completions while you type, a chat that explains code, or an agent that takes a task and comes back with a diff.
3. Who pays: you, or an employer with rules about where code can be sent.

Someone who mostly wants fast autocomplete in IntelliJ needs something different from someone who wants to hand off a refactor from the terminal and review it later. If you skip this step, it's easy to end up paying for the wrong tool.

The five AI coding assistants at a glance

Prices as of November 2026, in US dollars, before tax.

What is each AI coding assistant best at?

GitHub Copilot

Copilot works in a long list of editors. GitHub lists VS Code, Visual Studio, JetBrains IDEs, Xcode, Neovim, Eclipse and Zed, plus a CLI and the GitHub website. If your team already lives in GitHub issues and pull requests, Copilot can take an issue and return a pull request, and Pro includes pull request code review.

Chat and agent work is billed through credits. As of November 2026, each paid plan includes a monthly credit allowance: $15 on Pro ($10 a month), $70 on Pro+ ($39) and $200 on Max ($100). Heavier models use more credits per request. Completions and next edit suggestions don't use credits and stay unlimited on paid plans. When the allowance runs out, you can wait for the monthly reset, set a dollar budget for extra usage, or switch to a lighter model.

Check one setting early. GitHub says it may use prompts and suggestions from individual Free, Pro and Pro+ users to train models, and you can opt out in your Copilot settings.

Cursor

Cursor is a full editor built around AI, with its agent, cloud agents and Bugbot code review in one product. As of November 2026 its plans are sold as a bundle with xAI's Grok and X subscriptions. The $8 Plus plan includes Grok models only. Premium ($30), Super ($100) and Ultra ($200) add cloud agents and AI code reviews, and they let you use other providers' models, billed separately at API list price. A month of heavy Claude or GPT use inside Cursor can therefore cost far more than the sticker price, so set a spending cap. Cursor says that with Privacy Mode on, your code is not used for training by Cursor or its model providers.

Claude Code

Claude Code is Anthropic's coding agent, best known for running in the terminal. As of November 2026 it comes with Claude Pro ($20 a month, or $17 a month billed yearly), Max ($100, five times Pro usage) and Ultra ($200, ten times Pro's weekly usage and no five-hour limit for verified accounts). Max and Ultra also include monthly API credits, $100 and $200 respectively. The free plan doesn't include Claude Code.

Usage is shared with your normal Claude chats, so a long refactor in the afternoon eats into the same allowance as your questions in the browser. When you hit a limit you can wait for the reset, upgrade, or turn on usage credits, which are billed at standard API rates.

OpenAI Codex

Codex comes with ChatGPT plans. As of November 2026, Free and Go ($8 a month) get Codex in the desktop app for lighter tasks, subject to rollout. Plus ($20) gets Codex on the web, in the CLI, in the IDE extension and on iOS, plus automatic code review and a Slack integration. Pro comes in $100, $200 and $500 tiers. You can also run the CLI with an API key and pay API rates, which suits CI pipelines.

OpenAI publishes rough ranges for how many local messages you get per five hours on each model. The ranges are wide, so check your usage dashboard during the trial instead of planning around the estimate.

Google Antigravity

Antigravity is Google's agent-first IDE, with a CLI and extensions for other editors. The individual plan is free, with unlimited tab completions and basic weekly rate limits. Google AI Pro and Ultra subscriptions raise those limits and add a flexible AI credit pool.

If you used Gemini Code Assist or Gemini CLI on a personal Google account, Google stopped serving those consumer tiers on June 18, 2026 and points users to Antigravity instead. It's a good reminder not to build your whole workflow around one vendor's plugin.

How do you test an AI coding assistant on your own code?

A feature list can't tell you which tool understands your codebase. A short trial on real work can.

Week 1: same tasks, two tools

Pick two finalists and give each the same five tasks from your real backlog:

• Fix a small bug with a failing test.
• Add an endpoint or component that follows an existing pattern.
• Write tests for a module with poor coverage.
• Explain an unfamiliar part of the codebase.
• Do a cross-file rename or refactor.

For each task, note whether the result compiled, whether tests passed, and how much you had to rewrite. A simple spreadsheet is enough.

On my Java and backend projects it's helped me understand errors and save time, but I still check the code myself.

Week 2: check the money and the limits

Use your favorite of the two as your only assistant for a normal working week. At the end, open its usage page and ask yourself: Did I hit a limit? How many credits did agent sessions use? Would the next plan up have been worth it?

Metered plans tend to surprise people here. GitHub's own FAQ says a quick question to a lightweight model costs a fraction of a credit, while a long agent session on a frontier model across many files costs much more.

Do AI coding assistants train on your code?

It depends on the plan and your settings. If the code belongs to an employer, ask before you install anything. Team plans add admin controls. Claude Team and ChatGPT Business both say they don't train on business data by default, and GitHub says its extra training use of Copilot data applies to individual plans. On a personal account, find the training opt-out or privacy toggle on day one.

Which AI coding assistant should you pick?

• For unlimited completions across many IDEs at a low flat fee, start with Copilot Pro.
• For a terminal agent that handles multi-file changes, and you already pay for Claude, try Claude Code.
• If you already pay for ChatGPT Plus, test Codex before buying anything else.
• For a free agent IDE to experiment with, try Antigravity.
• For an AI-first editor, and you're fine with usage-based billing for non-Grok models, look at Cursor Premium.

Whichever you choose, review every diff. GitHub's FAQ warns that Copilot can suggest insecure code, the same caution applies to the others, and your name is on the commit.

FAQ

Is GitHub Copilot free?

Yes, there is a free tier. As of November 2026, Copilot Free includes 2,000 completions a month, limited chat and agent use, and the Copilot CLI, with no credit card required. Pro costs $10 a month and adds unlimited completions, pull request code review and a $15 monthly credit allowance.

Is Claude Code free?

No. Claude Code isn't part of Anthropic's free plan. As of November 2026 it comes with Claude Pro at $20 a month, or $17 a month billed yearly, and with the Max ($100) and Ultra ($200) plans. Claude Code sessions draw from the same usage limits as your regular Claude chats.

Is a free AI coding assistant enough for a student?

For learning, often yes. Copilot Free gives you 2,000 completions a month, and Google Antigravity's free tier includes unlimited tab completions with basic weekly rate limits. Verified students can also get the GitHub Copilot Student plan. Upgrade only when you keep hitting limits on real project work.

Do AI coding assistants work in IntelliJ IDEA?

Some do. GitHub Copilot supports JetBrains IDEs, including IntelliJ IDEA, and its plans page lists agent mode for JetBrains. Google Antigravity offers extensions for other editors. Cursor is a separate editor, so it replaces IntelliJ instead of plugging into it. Check each vendor's docs for your IDE version before paying.

Why does my AI coding assistant bill change every month?

Most plans now meter chat and agent work with credits or tokens, and bigger models use more per request. GitHub says a quick question to a lightweight model costs a fraction of a credit, while a long agent session on a frontier model across many files costs much more. Set a spending cap where you can.

Sources

Pricing pages checked November 2026.

• GitHub Copilot plans and pricing: https://github.com/features/copilot/plans
• Cursor pricing: https://cursor.com/pricing
• Claude plans and pricing: https://claude.com/pricing
• OpenAI Codex pricing: https://developers.openai.com/codex/pricing
• Google Antigravity pricing: https://antigravity.google/pricing
• Gemini Code Assist consumer account deprecation: https://codeassist.google/