AI coding agents compared, November 2026

All coding agents at a glance

Monthly US prices for individual plans, from each vendor's pricing page on 4 November 2026.

Codex Free and Go use one model in the desktop app only. Cursor Plus runs Cursor's own and Grok models only; third-party models start on Premium ($30).

What changed in 2026

• Google Jules is shutting down. Google's changelog (29 Oct) says sessions can be finished until 11 November 2026; Jules is then read-only until it closes on 31 December. Google points users to Antigravity.
• Claude's Max 20x plan is now Ultra (30 Oct), still $200 a month, with 10x Pro usage per week and no five-hour limit for verified accounts.
• Cursor joined xAI's X Subscription (12 Oct). New plans are Plus $8, Premium $30, Super $100 and Ultra $200, with a weekly usage pool shared with Grok. Old Pro and Pro+ plans keep working as legacy plans.
• Windsurf became Devin Desktop (2 Jun) and now shares Devin's plans. Windsurf for JetBrains continues as a separate product.
• GitHub Copilot moved to AI credits (1 Jun). Each credit is $0.01; completions are free, while agent mode, code review, the cloud agent and the CLI use credits. Copilot Max ($100) launched the same day.
• Devin cut its team price (14 Apr). The $500-a-month Team plan became Teams at $80 a month plus $40 per developer seat, and the old pay-as-you-go plan became Free.

Which AI coding agent is cheapest?

• Free with your own key: Cline is open source. You pay your model provider directly, at their API prices.
• Free plan: Google Antigravity has a free individual plan with weekly limits and unlimited Tab completions. GitHub Copilot Free includes 2,000 completions a month.
• Cheapest paid: Google AI Plus ($4.99) gives a limited Antigravity plan. Cursor Plus ($8) is cheap but limited to Cursor's own and Grok models. Copilot Pro ($10) includes 1,500 AI credits and premium models.
• Already paying for a chatbot: ChatGPT Plus includes Codex and Claude Pro includes Claude Code, both at $20, so there may be nothing extra to buy.

How is usage billed?

Two $20 plans can give very different amounts of work. This is how each vendor meters it:

Terminal, editor or cloud agent?

• Terminal-first: Claude Code and OpenAI Codex run in your shell and also ship IDE extensions and desktop apps.
• Editor-first: Cursor, Devin Desktop, Antigravity and Kiro are full editors with an agent built in. Kiro writes a spec (requirements, design, tasks) before it codes.
• Inside your current IDE: GitHub Copilot and Cline are extensions for VS Code and JetBrains.
• Background agents: Devin, Augment Code, Factory, Codex Cloud and Copilot's cloud agent take a task and return a pull request.

For head-to-head details, see Claude Code vs Cursor, Claude Code vs Codex and Cursor vs GitHub Copilot, or browse all coding AI tools.

Team plans

What about benchmarks?

There is no current neutral leaderboard. The newest entry in the official SWE-bench Verified data is from February 2026, so it predates every model these agents now use. Vendor-published scores test models, not the agent around them, and each vendor picks its own settings. Treat them as claims and try two agents on your own code instead: most have a free plan or a $20 tier you can cancel.

How we checked

Every plan and price comes from the vendor's pricing page or documentation, opened on 4 November 2026. None of the vendors publish exact quota numbers except where stated. For subscription prices of the chat apps that include these agents, see AI subscription prices.