GitHub Copilot vs Claude Code vs Codex: Desktop App, CLI and Price
All three now have a desktop app, a CLI, IDE extensions and a cloud agent, so billing decides. Paid plans start at $8 (Codex), $10 (Copilot) and $20 (Claude Code), compared by surface.

When this comparison first ran in May 2026, the three tools lived on different surfaces: GitHub Copilot had just announced a desktop app, while Claude Code and Codex were terminal agents. That difference is gone. As of November 2026, all three ship a terminal CLI, a desktop app, IDE extensions and cloud sessions. The factor that now settles the choice is billing: which subscription you already pay for, and whether you want metered credits or a flat plan with usage limits.
Surface and price, side by side
As of 2026-11-19. Each cell links to the vendor page it comes from.
| GitHub Copilot | Claude Code | Codex | |
|---|---|---|---|
| Desktop app | Yes: the GitHub Copilot app, built on Copilot CLI. macOS, Windows, Linux. All plans, including Free | Yes: the Code tab in the Claude desktop app. macOS, Windows; Linux (Ubuntu, Debian) in beta. Paid plan required | Yes: Codex inside the ChatGPT desktop app. macOS, Windows, Linux |
| CLI | Copilot CLI: Linux, macOS, Windows (PowerShell or WSL). All plans | Claude Code CLI: macOS, Linux, WSL, Windows | Codex CLI: Plus and higher, or an API key |
| IDE | Copilot in IDEs on all plans | VS Code (and Cursor), JetBrains | IDE extension: Plus and higher, or an API key |
| Cloud agent | ”Copilot in the cloud” on paid plans, not Free | Claude Code on the web (claude.ai/code) and the Claude iOS and Android app | Codex Cloud and Codex on the web; not with an API key |
| Browser extension | None listed in GitHub’s plan docs; Copilot Chat runs on GitHub.com | Claude in Chrome, driven from the CLI or VS Code with claude --chrome. Chrome, Edge and other Chromium browsers | ChatGPT browser extension, used from Codex chats in the ChatGPT desktop app. Chrome, Edge, Brave, Opera, Vivaldi |
| Free tier | Yes: app, CLI and Copilot in IDEs; 2,000 inline suggestions a month | No: the Free plan excludes Claude Code | Yes: GPT-6 Luna in the desktop app only, subject to rollout |
| Cheapest paid plan | Pro, $10/month (1,500 AI credits) | Pro, $20/month ($17 billed annually) | Go, $8/month (desktop app only); Plus, $20/month for CLI, IDE and web |
| Higher individual plans | Pro+ $39, Max $100 | Max $100, Ultra $200 | Pro $100, $200 or $500 |
| Team / business seat | Business $19, Enterprise $39 per seat | Team standard seat $25 ($20 annual), premium $125 ($100 annual); Enterprise $20 per seat annual plus usage | Business $25 per user ($20 annual) |
| Billing model | AI credits charged at each model’s API rate | Flat plan with usage limits (5-hour and weekly; no 5-hour limit on Ultra), or API pay-as-you-go | Flat plan with 5-hour limits (none on Pro currently), or API pay-as-you-go |
| Models | Anthropic, OpenAI, Google, xAI and others, plus Copilot Fusion routing | Anthropic Claude only | OpenAI only (or any model your API key can reach) |
| Best for | Teams that review AI work in GitHub, or want several vendors’ models | Developers already on a Claude plan | Developers already on a ChatGPT plan |
Quick answers
Does GitHub Copilot have a desktop app like Claude?
Yes. GitHub calls it the GitHub Copilot app: “a desktop application purpose-built for agent-driven development” (GitHub docs). It runs on macOS, Linux and Windows and is available on every Copilot plan. Like the Code tab in the Claude desktop app, it runs several agent sessions side by side. It can also create, review and merge pull requests and show CI results; the Claude desktop app shows CI status and can auto-merge PRs too.
GitHub Copilot app vs Copilot CLI: what is the difference?
The app “is built on GitHub Copilot CLI” (GitHub docs), so both run the same agent. The CLI lets you “use Copilot directly from your terminal” (CLI docs). The app adds a window for parallel sessions, each with its own git worktree and branch, plus pull request and CI views. Both are included on all Copilot plans (plans).
Does Copilot have something like Claude Code or Codex?
Yes, in two ways. Copilot CLI and the Copilot app are GitHub’s own terminal and desktop coding agents, the same shape as Claude Code and Codex. Separately, Copilot can run Anthropic’s Claude agent and OpenAI’s Codex agent as third-party agents on Pro, Pro+, Max, Business and Enterprise. That feature is in public preview.
Codex desktop vs Claude Code desktop?
Both live inside the vendor’s main desktop app. Codex lives in the ChatGPT desktop app (macOS, Windows, Linux), and its Free and Go plans give limited access there (pricing). Claude Code is the Code tab of the Claude desktop app (macOS and Windows, Linux in beta), and it needs a paid Claude plan. Claude’s app can review diffs visually, run sessions side by side, schedule recurring tasks and start cloud sessions.
Is there a Codex Chrome extension?
Yes. OpenAI’s ChatGPT browser extension lets a Codex chat in the ChatGPT desktop app work in Chrome, Edge, Brave, Opera or Vivaldi. By default it asks before it interacts with each new website. Claude Code’s equivalent is Claude in Chrome, which requires a Pro, Max, Ultra, Team or Enterprise plan and does not work with API-key sign-in. GitHub’s plan docs list no Copilot browser extension.
Why billing decides this matchup now
The three products converged on the same shape. Each one can read your repository, edit files across a project, run commands, work in parallel sessions and hand work off to the cloud. The Copilot app went generally available on June 17, 2026, and reached every Copilot plan on July 7. Claude Code runs in the Claude desktop app as well as the terminal. Codex now lives inside the ChatGPT desktop app next to its open-source CLI. Picking by surface no longer separates them.
What still separates them is how you pay, and the three billing models behave differently once you use an agent all day:
- Copilot meters tokens. Since June 1, 2026, every Copilot plan bills in GitHub AI Credits. Usage is “calculated based on token consumption, including input, output, and cached tokens” at “the listed API rates for each model.” Your plan includes a monthly credit allowance, and you can buy more. Code completions and next edit suggestions do not consume credits.
- Claude Code draws on a flat plan. A Claude subscription covers Claude Code with usage limits that reset every 5 hours and every week. How fast you reach a limit depends on the model, the chat length and the file sizes. You can also connect an API key and pay per token instead.
- Codex draws on a ChatGPT plan. ChatGPT plans include Codex with per-5-hour message limits that vary by model, and weekly limits can also apply. An API key option bills at API rates instead.
The practical effect: on Copilot, a long agent session on a frontier model draws down your allowance at per-token prices, and the bill grows with the work. On Claude Code and Codex, the monthly price is fixed and the constraint is the usage window. If you already pay for one of these subscriptions, the agent that comes with it costs you nothing extra to try. That alone decides the choice for most developers.
GitHub Copilot: the app, the CLI and metered credits
The GitHub Copilot app is, in GitHub’s words, “a desktop application purpose-built for agent-driven development,” built on Copilot CLI (docs). It runs on macOS, Linux and Windows (with WSL 2 support). Each parallel session gets its own git worktree and branch. Sessions run locally, where you can configure a sandbox, or in a cloud sandbox that you can follow from any Copilot client. You choose between Interactive, Plan (you approve the agent’s plan) and Autopilot (fully autonomous) modes. You can create, review and merge pull requests and check CI results inside the app. Users without a Copilot plan can still run it by bringing their own model key.
Plans (GitHub docs, plan announcement):
- Free: limited chat and agent usage, 2,000 completions per month
- Pro, $10/month: 1,500 credits ($10 base + $5 flex)
- Pro+, $39/month: 7,000 credits ($39 base + $31 flex)
- Max, $100/month: 20,000 credits ($100 base + $100 flex)
- Business, $19/seat: 1,900 credits per user; Enterprise, $39/seat: 3,900 credits per user
GitHub says the flex portion “will vary over time” as model prices change. Since October 28, 2026, GitHub Enterprise customers working with an account team get Copilot with their license and pay for AI usage from the first request, with no per-seat allowance.
Models. Copilot’s supported models list covers Anthropic (Claude Opus 5.5, Sonnet 5.5, Fable 5.2 and others), OpenAI (GPT-6.2 Astra and Sol, GPT-6.1 Sol and Terra, GPT-6 Astra, Sol and Luna, and GPT-5.x models), Google (Gemini 4 and 3.8), xAI (Grok 4.x), Moonshot AI and Microsoft. Instead of picking a model, you can choose Copilot Fusion, generally available since October 28. You pick an effort level (Standard, High or Max) and Copilot routes or combines models for you. Our Copilot Fusion explainer covers how that affects your bill.
Claude and Codex inside Copilot. Copilot can also run Anthropic’s Claude agent and OpenAI’s Codex agent as third-party agents, in public preview. You start them from a repository’s Agents tab, an issue, a pull request comment, GitHub Mobile or VS Code. They are powered by your Copilot subscription and consume GitHub Actions minutes and AI credits. If your team wants Claude and Codex on one GitHub bill, this is the route.
Claude Code: flat plans, Claude models only
Claude Code reads your codebase, edits files, runs commands and works on the terminal, VS Code, JetBrains, a desktop app and the web. The desktop app is the Code tab in the Claude app, where you review diffs visually, run sessions side by side, schedule recurring tasks and start cloud sessions. It requires a paid subscription. The CLI stays composable: you can pipe logs into claude -p, run it in CI, or coordinate it with other terminal agents.
Approvals. With Claude Code v2.1.283 or later, interactive terminal and VS Code sessions start in auto mode (permission modes). In auto mode, “a separate classifier model reviews actions before they run, blocking anything that escalates beyond your request, targets unrecognized infrastructure, or appears driven by hostile content Claude read.” Anthropic warns that auto mode “reduces permission prompts but does not guarantee safety.” Switch to Manual mode if you want to approve edits and shell commands yourself: there, Claude Code “stops and asks you before most actions that edit files, run shell commands, or reach the network.” Manual is also the starting mode for claude -p and Agent SDK runs. Other modes accept file edits automatically or plan only.
Plans (Claude pricing, Max and Ultra):
- Free: no Claude Code
- Pro: $20/month, or $17/month billed annually
- Max: $100/month, 5x Pro usage per 5 hours and per week
- Ultra: $200/month, 10x Pro usage per week
- Team: standard seat $25/month ($20 annual), premium seat $125/month ($100 annual)
- Enterprise: $20/seat/month billed annually, plus usage at API rates
Models. Claude Code runs Anthropic’s Claude models only. The current lineup is Claude Fable 5.2, Opus 5.5, Sonnet 5.5 and Haiku 5.5, all with a 1M-token context window. The CLI can also connect through Amazon Bedrock, Google Cloud or Microsoft Foundry, but those still serve Claude models.
Codex: ChatGPT plans, OpenAI models, open-source CLI
Codex CLI is “a coding agent from OpenAI that runs locally on your computer,” and its source code is on GitHub under the Apache-2.0 license. When an agent does something unexpected, you can read the code that made the decision. The same agent is available in the ChatGPT desktop app, the ChatGPT mobile app, ChatGPT on the web, the IDE extension and Codex Cloud. You sign in with your ChatGPT account, or set up an API key. In the CLI, the /permissions command sets “when Codex can edit files or run commands without asking” (Codex CLI docs).
Plans (Codex pricing):
- Free ($0) and Go ($8/month): GPT-6 Luna at standard speed in the desktop app, subject to rollout
- Plus, $20/month: GPT-6.1 Sol, GPT-6 Luna, GPT-6.2 Astra and GPT-6.2 Sol, with per-5-hour local message limits
- Pro, $100, $200 or $500/month: no five-hour limit currently
- Business: $20/user/month billed annually, $25 billed monthly
- API key: pay-as-you-go at API rates, without cloud features such as GitHub code review
Models. OpenAI released GPT-6.2 Astra and GPT-6.2 Sol on November 17, 2026. Astra is the frontier model for the hardest projects, and Sol is the efficient model for everyday work. Both are available in the Codex app and CLI on Plus, Pro and Business; on Enterprise and Edu they are off by default, and an administrator must turn them on (Codex pricing). On Plus, OpenAI estimates 17 to 180 local messages per 5 hours with GPT-6.2 Sol, so heavy users are likely to hit the window.
Choosing for your daily workflow
Start from the subscription you already pay for, because the bundled agent is free to try:
- You pay for ChatGPT Plus or Pro: use Codex. It comes with the plan, runs OpenAI’s newest models including GPT-6.2, and the CLI is open source.
- You pay for Claude Pro, Max or Ultra: use Claude Code. It comes with the plan and runs Anthropic’s newest models directly.
- Your company pays for Copilot, or you want one bill for several vendors’ models: use Copilot. The app, the CLI, Fusion and the third-party Claude and Codex agents all draw from the same credits.
- You pay for none of them: choose by billing model. Codex Go ($8) and Copilot Pro ($10) are the cheapest paid entries; Claude Pro ($20) is the dearest. Codex Free and Go cover only the desktop app, so the Codex CLI and IDE extension start at Plus ($20). Codex and Copilot also have free options, and Claude Code has none.
Runner-up factor: where your reviews happen. If your team audits AI work through GitHub issues, pull requests and Actions, Copilot puts the agent’s output where reviewers already look, and it can run Claude and Codex agents on the same surface. Claude Code also opens pull requests and runs in GitHub Actions, and Codex offers GitHub code review on ChatGPT plans, but GitHub is Copilot’s home ground.
What flips the answer. If your agent usage is heavy and steady, compare the flat price with the metered one. A Copilot plan converts your allowance to tokens at each model’s API rate, so a month of long agent sessions on frontier models can outrun the included credits. A Claude or ChatGPT plan at the same price keeps the bill fixed and limits usage per window instead. If predictable cost matters more than model choice, a flat plan wins even for a GitHub-centric team.
What this article did not test
- No hands-on benchmark of the three agents. Statements about features, approvals and limits come from vendor documentation and changelogs linked inline.
- No measurement of how many tasks a given plan supports. Usage-limit figures are vendor estimates and change with the model and the task.
- No comparison of output quality between models. Vendor benchmark claims, such as GitHub’s Fusion score and OpenAI’s GPT-6.2 rankings, are reported by the vendors themselves.
Surfaces and plan prices were re-checked against vendor pages on November 19, 2026; model lineups were last checked on November 18, 2026. All three vendors changed pricing at least once in 2026, so confirm the current figures before you buy.
OpenCode
Another terminal-native option: OpenCode runs in your shell, edits files, runs tests, and works with whichever editor you already use. Worth testing if you like Claude Code's shape but want a different model and billing model.
Free tier available
Affiliate link · We earn a commission at no cost to you.
There is still a category these three do not cover: the editor whose writing surface is built around the agent. Cursor occupies that niche. If a third window feels like one window too many, an AI-native IDE is the alternative.
FAQ
Does the GitHub Copilot desktop app still exist?
Does the GitHub Copilot app replace the IDE extension?
Can I use Claude models or Claude Code inside Copilot?
Is Claude Code included in the Claude free plan?
Which tool has the smallest risk of an agent breaking my repository?
Three tools, one converged shape. Pick the one your existing subscription already pays for, and switch only when model choice or GitHub-native review is worth a second bill.
Tools used in this review
Some links above are affiliate links. We may earn a commission if you sign up. See our disclosure for details.
Related reading
2026-09-02
DeepSeek MLA: 70 GB of KV Cache at 1M Tokens
No DeepSeek-V4 config is public yet. The V3 one is, and its KV-cache math tells you what a million-token window actually costs in GPU memory.
2026-08-13
AI Agent Auth: API Keys vs Device Flow vs Scoped Tokens
Three credential models for non-human callers: static keys, the OAuth 2.0 device grant, and short-lived scoped tokens - and when each one fits.
2026-08-12
Writing a Spec an AI Agent Can Actually Execute
Ground truth files, an interface contract, one acceptance command, and explicit out-of-bounds rules -- so the agent runs end to end without babysitting.
2026-07-16
Multi-Agent Terminal Workflows with OpenCode and Aider
How we split work across OpenCode, Claude Code, and Aider in one terminal without losing track of which agent changed what.
2026-07-16
OpenCode First Project Setup: Install to First Passing Test
Configure providers, context files, and project conventions so the agent produces useful output on an existing codebase from day one.
Get the best tools, weekly
One email every Friday. No spam, unsubscribe anytime.