Skip to content
TBSI.

Start here

  • LLM LeaderboardCompare frontier models on intelligence, price and speedTool
  • What is the best superintelligence?The best frontier model for each jobModels
  • Frontier Gap ReportHow far apart the leading labs areReport
  • MethodologyHow we verify sources, data and claimsAbout

Best AI for Coding (2026): Claude Code, GitHub Copilot, Cursor and Codex Compared

The best model and the best tool for programming, from vendor benchmark tables and official pricing pages read on 2026-11-17

By TheBestSuperintelligence Editorial TeamUpdated

◈ Executive SummaryCited · Documentation

What is the best AI for coding right now?

The best AI for coding in 2026 is Claude Code running Claude Opus 5.5. Opus 5.5 has the top Terminal-Bench 4.0 score in both Anthropic's and Google's tables (66.4%), and Claude Code comes with Claude Pro at $20 a month. For an editor, use GitHub Copilot.

◈Key Takeaways (4)
  • •Best model: Claude Opus 5.5, the top Terminal-Bench 4.0 score in both Anthropic's and Google's tables (66.4%).
  • •Best tool for agent work: Claude Code, included with Claude Pro ($20 a month).
  • •Best in the editor: GitHub Copilot in VS Code, from free to $10 a month for Pro.
  • •Gemini CLI stopped serving free individual users on June 18, 2026; Google moved them to Antigravity CLI.
☰On this page (11 sections)
  1. 01.Which AI for coding fits your job?
  2. 02.Best AI model for coding: Claude Opus 5.5
  3. 03.Best AI coding assistant for agent work: Claude Code
  4. 04.Best for coding inside your editor: GitHub Copilot in VS Code
  5. 05.Best AI-first editor: Cursor
  6. 06.Best if you already pay for ChatGPT: OpenAI Codex
  7. 07.What happened to Gemini CLI?
  8. 08.Is there a free AI for coding?
  9. 09.Claude Code vs Cursor vs GitHub Copilot: how to choose
  10. 10.Limits of this guide
  11. 11.Sources & Further Reading

Which AI for coding fits your job?

The best AI for coding is two choices: a model and a tool. Our pick is Claude Code running Claude Opus 5.5. Opus 5.5 holds the top Terminal-Bench 4.0 score in both Anthropic's own table and Google's rival table (66.4%), and Claude Code runs it wherever you work, from the terminal to VS Code. If you would rather stay in your editor, pick GitHub Copilot.

Your job Pick Why
Agent work across a repository Claude Code + Opus 5.5 Top Terminal-Bench 4.0 score; terminal, IDE, desktop and web
Coding inside VS Code GitHub Copilot Free tier; Claude, GPT and Gemini models
An AI-first editor Cursor Agents and Projects built into the editor
You already pay for ChatGPT OpenAI Codex Included in ChatGPT Plus at $20 a month
Paying nothing Copilot Free 2,000 completions a month

Best AI model for coding: Claude Opus 5.5

Claude Opus 5.5 is the model to start with. Anthropic announced it on 2026-09-22 at $4 input and $20 output per 1M tokens. On Terminal-Bench 4.0, which tests agents on multi-step jobs in a real command line, it scores 66.4% in Anthropic's table and the same 66.4% in Google's Gemini 4 Argon model card. Anthropic also reports 57.8% on CursorBench 4.0, built from real Cursor sessions.

Here are the coding rows from Google's card, the one table that lists all three frontier models side by side (a rival's table, so read it with the same care):

Benchmark (vendor-reported) Claude Opus 5.5 GPT-6 Astra Gemini 4 Argon
Terminal-Bench 4.0 66.4% 58.2% 57.4%
DeepSWE v1.1 74.2% 74.1% 77.9%
FrontierSWE v2 62.3% 65.5% 55.0%

The lead is not total. Gemini 4 Argon tops DeepSWE and GPT-6 Astra tops FrontierSWE. OpenAI calls Astra "the best model for software engineering to date", but its own table compares Astra with Claude Fable 5.1 and Opus 5, not Opus 5.5. None of the three launch pages we read reports SWE-bench Verified, the benchmark most "best AI for coding" lists still quote, so we do not rank on it.

Our take: Terminal-Bench is closest to what coding agents actually do (run commands, read errors, try again), and it is the one row where both rival tables agree on the leader by a wide margin. That is why Opus 5.5 is our default. For the full cross-vendor picture, see our best AI model guide.

Best AI coding assistant for agent work: Claude Code

Claude Code is Anthropic's agentic coding tool. Its documentation says it "reads your codebase, edits files, runs commands, and integrates with your development tools". It runs in the terminal, as a VS Code extension (which also installs in Cursor), as a JetBrains plugin, in a desktop app and in the browser at claude.ai/code, and all of them share the same project settings.

The details that matter for daily use:

  • Price: included in Claude Pro at $20 a month ($17 a month billed yearly) and in Max from $100 a month. The Free plan does not include it. An Anthropic Console (API) account also works.
  • Project rules: a CLAUDE.md file holds your standards; Claude Code can read an existing AGENTS.md instead.
  • Speed: Opus 5.5 has a fast mode in Claude Code at up to 2.5 times the speed, billed at $8 input and $40 output per 1M tokens on the API.
  • Automation: subagents in parallel, hooks, scheduled tasks in the cloud, and GitHub Actions or GitLab CI/CD for code review.

Best for: multi-file changes, bug hunts and refactors you hand off and review. Weakness: it is built around one vendor's models, so if you want to switch between Claude, GPT and Gemini per task, an editor like Copilot is more flexible. Prompts that work well with it are in our Claude Code prompts.

Best for coding inside your editor: GitHub Copilot in VS Code

GitHub Copilot is the pick if you want AI inside the editor you already use, with a choice of models. GitHub's supported-models page lists Claude Opus 5.5, Claude Sonnet 5.5, GPT-6 Astra, GPT-6.1 Sol and Gemini 4 Argon as generally available. In VS Code it can also run Anthropic's and OpenAI's own agent harnesses, which we explain in our VS Code AI agents guide.

Plan US price What GitHub lists
Free $0 2,000 completions a month; limited chat and agent usage; no coding agent
Pro $10 per user a month Unlimited completions; $15 of monthly AI credits; coding agent
Pro+ $39 per user a month About 4 times Pro's usage; premium models including Opus
Max $100 per user a month About 2.9 times Pro+ usage; priority access to new models

GitHub defines one AI credit as $0.01. Business and Enterprise prices are not on the plans page.

Best for: everyday coding with completions plus an agent, and teams on GitHub. Weakness: the Opus model on its own needs Pro+ at $39, almost twice the price of Claude Pro.

Best AI-first editor: Cursor

Cursor is a full editor built around agents: an agent pane, an Agents Window, and Projects, where a coordinator agent plans work and delegates it to other agents. Our Cursor Agent guide covers those features in depth.

The pricing changed shape. Cursor's pricing page, read 2026-11-17, lists Free (limited usage), Plus at $8 a month, Premium at $30, Super at $100 and Ultra at $200, with Teams from $30 per user. Plus offers only Grok models. Premium adds cloud agents and third-party models at API list price. The page says subscriptions are sold through X Subscription Corp.

Our take: Cursor is now worth it at Premium or above. On Plus you cannot pick Claude, GPT or Gemini models, and we found no cross-vendor coding benchmark for the Grok models on the page.

Best if you already pay for ChatGPT: OpenAI Codex

OpenAI's Codex page says Codex "is included in your ChatGPT plan": Plus at $20 a month "for focused coding sessions each week", Pro at $100 a month for full workdays, and Business at $20 per user a month (two or more seats, billed yearly). It runs in ChatGPT, in an IDE extension, in the terminal as Codex CLI, and in the cloud, where tasks continue after you close your laptop. It also reviews pull requests.

GPT-6 Astra is the model to use in it. OpenAI reports 57.9% on Terminal-Bench 4.0 and 74.1% on DeepSWE v1.1, and says Astra in Codex can keep notes across context windows instead of summarising them away (an experimental setting for now).

Best for: ChatGPT subscribers who want an agent without a second bill. We could not confirm whether the free ChatGPT plan includes Codex.

What happened to Gemini CLI?

Gemini CLI is no longer the free option it was. Google's developer blog (2026-05-19) says that on June 18, 2026, Gemini CLI stopped serving individual users on Google AI Pro and Ultra and those using it free through Gemini Code Assist for individuals. They were moved to Antigravity CLI, part of Google Antigravity, which Google says is "available to everyone". Gemini CLI itself remains open source (Apache 2.0) for Gemini Code Assist Standard and Enterprise customers and paid API keys.

If you use Gemini models for coding, the route now is Antigravity, or Gemini 4 Argon inside Copilot.

Is there a free AI for coding?

Yes, three worth trying, each with limits:

  • GitHub Copilot Free: 2,000 completions a month plus limited chat and agent use. The best free start.
  • Google AI Individual in Antigravity: $0 a month with "unlimited Tab completions", "unlimited Command requests" and "basic weekly rate limits". Google's pricing page does not name the models or the exact limits.
  • Cursor Free: "limited" usage, with no figures given.

Claude's Free plan and the free tier of Gemini CLI are not options: the first excludes Claude Code, and the second ended on June 18. For a free tool, good prompts matter more; ours are in best AI prompts for coding.

Claude Code vs Cursor vs GitHub Copilot: how to choose

Claude Code Cursor GitHub Copilot
What it is Agent that works across your repo AI-first editor AI inside your editor
Where it runs Terminal, IDEs, desktop, web Its own editor VS Code and other editors
Lowest paid plan $20 a month (Claude Pro) $8 (Grok only); $30 for other models $10 a month (Pro)
Models Claude Grok; others from Premium Claude, GPT, Gemini

Choose by where you want the work to happen. If you describe a task and review the result, Claude Code. If you type code and want help line by line, Copilot. If you want both in one editor and will pay $30, Cursor.

Limits of this guide

The benchmarks are the labs' own and were run with different settings, so treat small gaps as noise. Plan limits are described in credits or "limited" usage, not fixed request counts, so we cannot say how far each subscription goes on a real project. Model lineups in Copilot and Cursor change monthly.

Sources & Further Reading

Claim Source Date
Opus 5.5 announcement, $4/$20 price, Terminal-Bench 4.0 66.4%, CursorBench 4.0 57.8%, fast mode $8/$40 at up to 2.5x Anthropic: Introducing Claude Opus 5.5 2026-09-22, read 2026-11-17
Terminal-Bench 4.0 standard error ±2.6 points for Opus 5.5 Anthropic: Introducing Claude Opus 5.5, footnote 1 2026-09-22, read 2026-11-17
Terminal-Bench 4.0, DeepSWE v1.1 and FrontierSWE v2 for Opus 5.5, Astra and Argon Google DeepMind: Gemini 4 Argon model card Published 2026-10-13, read 2026-11-17
Astra "best model for software engineering", Terminal-Bench 4.0 57.9%, DeepSWE 74.1%, notes across context windows in Codex OpenAI: GPT-6 Astra Internet Archive copy of 2026-11-16, read 2026-11-17
Codex included in ChatGPT Plus $20, Pro $100, Business $20 per user; surfaces; PR review OpenAI: Codex Internet Archive copy of 2026-11-16, read 2026-11-17
What Claude Code does, surfaces, CLAUDE.md and AGENTS.md, subagents, hooks, scheduled tasks, CI Claude Code overview Read 2026-11-17
Claude Free, Pro ($20, $17 yearly) and Max ($100); Claude Code on Pro and Max only Claude pricing Read 2026-11-17
Copilot Free, Pro, Pro+ and Max prices, completions, credits, coding agent GitHub Copilot plans Read 2026-11-17
Opus 5.5, Sonnet 5.5, GPT-6 Astra, GPT-6.1 Sol, Gemini 4 Argon generally available in Copilot GitHub Docs: Supported AI models Read 2026-11-17
Cursor plans, Grok-only Plus, Premium third-party models, X Subscription Corp. Cursor pricing Read 2026-11-17
Gemini CLI moved to Antigravity CLI; June 18, 2026 cutoff; who keeps access Google for Developers: Transitioning Gemini CLI to Antigravity CLI 2026-05-19, read 2026-11-17
Gemini CLI open source (Apache 2.0); no free tier Gemini CLI on GitHub Read 2026-11-17
Google AI Individual $0 with Tab and Command requests, weekly limits Google Antigravity pricing Read 2026-11-17

Frequently asked questions

What is the best AI for coding right now?

The best AI for coding in 2026 is Claude Code running Claude Opus 5.5. Opus 5.5 has the top Terminal-Bench 4.0 score in both Anthropic's and Google's tables (66.4%), and Claude Code comes with Claude Pro at $20 a month. For an editor, use GitHub Copilot.

Is GitHub Copilot free?

Yes, with limits. GitHub's plans page lists Copilot Free at $0 with 2,000 completions a month and limited chat and agent usage. The coding agent that turns an assigned task into a pull request starts with Pro at $10 per user a month.

Can I use Claude Opus 5.5 inside GitHub Copilot?

GitHub's supported-models page lists Claude Opus 5.5 as generally available in Copilot, alongside GPT-6 Astra and Gemini 4 Argon. Which models you see depends on your Copilot plan and where you use Copilot, and GitHub's plans page reserves premium models including Opus for Pro+ and above.

Does Gemini CLI still work for free?

No. Google says Gemini CLI stopped serving free individual users and Google AI Pro and Ultra subscribers on June 18, 2026. They were moved to Antigravity CLI. Gemini CLI stays open source under Apache 2.0 for Gemini Code Assist Standard and Enterprise customers and paid API keys.

Why do vendor coding benchmarks disagree?

Each lab picks its rivals and run settings. OpenAI's GPT-6 Astra table reports 57.9% on Terminal-Bench 4.0 but leaves out Claude Opus 5.5; Google's table lists Astra at 58.2% and Opus 5.5 at 66.4%. Compare numbers only within one table, and test on your own repository.

How we sourced this. We read the vendors' benchmark tables, pricing pages and documentation on 2026-11-17 (OpenAI pages via Internet Archive copies). We did not test the tools ourselves. Methodology