Coding agents compared

How this comparison is laid out

This article compares four coding agents on the things you can check before you commit: how to install each one, which project instruction file it reads, which models it offers, how it is billed and what its licence says. Each tool gets its own section. The last section sums up which one fits which situation. Prices and model names change often, so confirm them on each vendor's pricing page before you buy.

Claude Code

Claude Code is an agentic coding tool. You can use it in your terminal, IDE, desktop app and browser. Most of these surfaces require a Claude subscription or an Anthropic Console account ^[1].

The quick start installs it on macOS, Linux and WSL with curl -fsSL https://claude.ai/install.sh | bash. On Windows PowerShell it uses irm https://claude.ai/install.ps1 | iex. It can also be installed with Homebrew, using brew install --cask claude-code, or with WinGet, using winget install Anthropic.ClaudeCode. Confirm the install with claude --version ^[2]. Before you begin, you need a Claude subscription (Pro, Max, Ultra, Team or Enterprise), a Claude Console account, or access through a supported cloud provider. Setting the ANTHROPIC_API_KEY environment variable can skip the login prompt ^[2].

Project instructions go in a CLAUDE.md file in the project root. Claude Code reads it at the start of every session. If a repository already has an AGENTS.md for other coding agents, Claude Code can read that file in place of CLAUDE.md ^[1].

Billing is by API token consumption. The cost guide gives a rough picture for enterprise deployments: an average of about $13 per developer per active day, $150-250 per developer per month, and costs below $30 per active day for 90% of users ^[3].

The default model setting resolves to Opus 5.5 on Pro, Max, Ultra, Team, Enterprise, the Anthropic API, Claude Platform on AWS, Amazon Bedrock and Google Cloud's Agent Platform. On Microsoft Foundry it resolves to Sonnet 4.5 ^[4].

The licence is proprietary. The repository's LICENSE.md reads that the software is copyright Anthropic PBC, with all rights reserved, and that use is subject to Anthropic's Commercial Terms of Service ^[5].

Codex CLI

Codex CLI is a coding agent from OpenAI. It runs locally on your computer. It has an IDE option for VS Code, Cursor and Windsurf, a desktop app that starts with codex app, and the cloud-based Codex Web at chatgpt.com/codex ^[6].

It installs on Mac or Linux with curl -fsSL https://chatgpt.com/codex/install.sh | sh. On Windows the command is powershell -ExecutionPolicy ByPass -c "irm https://chatgpt.com/codex/install.ps1 | iex". It can also be installed with npm install -g @openai/codex or brew install --cask codex ^[6]. Run codex and choose Sign in with ChatGPT to use it as part of a Plus, Pro, Business, Edu or Enterprise plan. An API key is also supported, with some extra setup ^[6]. The repository is licensed under the Apache-2.0 License ^[6].

Codex reads AGENTS.md files before it does any work. It starts from its home directory, ~/.codex or the directory named by CODEX_HOME, and it prefers AGENTS.override.md there. The combined size of these instructions is capped by project_doc_max_bytes, which defaults to 32 KiB ^[7].

The ChatGPT desktop app, the Codex CLI and the IDE extension share one config.toml. If no model is specified they use a recommended model. The official example sets model = "gpt-6.1-sol" ^[8].

The pricing page lists a Free plan at $0 a month. Go is $8 a month, Plus is $20 a month, and Pro starts at $100 a month with plans at $100, $200 or $500 a month in US dollars. Business costs $20 per user a month when billed annually, or $25 when billed monthly ^[9]. The Free plan includes GPT-6 Luna at Standard speed in the desktop app, subject to rollout, and it does not include image generation ^[9]. For Plus and Standard Business, the page gives estimates of local messages per five-hour period. For GPT-6.1 Sol the estimate is 15-160. The page says these are not fixed message limits, and that Pro plans have no five-hour limit ^[9].

Gemini CLI

The Gemini CLI README describes gemini as an open-source terminal agent. It is maintained for enterprise Gemini Code Assist workflows on Google Cloud. For Gemini Code Assist Standard and Enterprise licences, and for Code Assist for GitHub, the README marks it as Enterprise Code Assist only ^[10].

The README also says there is no free tier and no Google AI plan. Free individual access and Google AI Pro and Ultra moved to Antigravity CLI on June 18, 2026 ^[10]. If you are an individual looking for a free terminal agent, check this point first.

The installation guide recommends npm install -g @google/gemini-cli, brew install gemini-cli, sudo port install gemini-cli, or running it immediately with npx @google/gemini-cli. It requires Node.js 20.0.0 or newer ^[11]. The README says the project is fully open source under the Apache 2.0 License ^[10].

Context files use the default name GEMINI.md. They load hierarchically from ~/.gemini/GEMINI.md and from workspace directories. The file name can be changed with the context.fileName property in settings.json ^[12]. The --model flag, or -m, defaults to auto. That resolves to gemini-2.5-pro or gemini-3-pro-preview, depending on whether preview features are enabled ^[13].

The quotas and pricing page says that pay-as-you-go cost depends on the model and on token usage. It also says /stats model shows a snapshot of the current session's token usage and quota limits ^[14].

TRAE

TRAE is a development tool with deep AI integration. It is available as a desktop app, a web app and a mobile app. TRAE Desktop has an Agent window and an IDE window, and the IDE window offers IDE mode and SOLO mode ^[15].

To install it, go to the TRAE official website, click Download, and complete the installation of the package. By default, TRAE opens the IDE window at launch ^[16]. TRAE Desktop supports macOS 12.0 or later on Apple Silicon and Intel, Windows 10 and Windows 11 in 64-bit, and Linux .deb packages for Ubuntu 20.04+ and Debian 10+, and .rpm packages for Fedora 42 and RHEL 9.x ^[16].

Project rules live in the .trae/rules/ directory under the project path. Global rules live in ~/.trae/user_rules on macOS and Linux, or in %userprofile%/.trae/user_rules on Windows ^[18]. TRAE can also reuse root-level AGENTS.md, CLAUDE.md and CLAUDE.local.md files. The docs describe AGENTS.md as a cross-tool convention, and say TRAE is compatible with the Claude Code files ^[18].

The models page lists built-in models that are available after upgrading to the new plan. They include GPT-6.1-Sol, GPT-6-Astra, GPT-6-Sol, GPT-6-Luna, GPT-5.5, GLM-5.2, DeepSeek-V4-Flash, Kimi-K2.7-Code and Kimi-K3 ^[17].

TRAE offers five plans: Free, Lite, Pro, Pro+ and Ultra. Their monthly prices are $0, $8, $20, $60 and $200 ^[19]. Free plan users get a limited Dollar Usage allowance each month. Once it is used up, AI chat is unavailable for the rest of the month ^[20]. On the Free plan, model selection is Auto mode only. Lite, Pro, Pro+ and Ultra offer both Auto mode and manual model selection ^[19]. In Auto mode the system picks a built-in model by weighing response speed, performance and resource use. After you enable Auto mode, the system does not use custom models ^[21].

Which one fits which situation

Read these as summaries of what each tool's documents describe, not as a verdict on code quality.

• Claude Code suits people who want one agent across terminal, IDE, desktop and browser, and who are comfortable with per-token billing or a Claude subscription. Its licence is proprietary ^[1][5].
• Codex CLI suits people who already have a ChatGPT plan, or who want a local agent with an IDE option and a cloud counterpart. Its licence is Apache-2.0 ^[6].
• Gemini CLI is open source under Apache 2.0. Its README places it in enterprise Code Assist workflows, and free individual access has moved to Antigravity CLI. Individuals should check that before they plan around it ^[10].
• TRAE suits people who want an IDE with a separate Agent window. On the Free plan, it offers Auto mode only and a limited monthly allowance ^[15][19].

If your repository already has a CLAUDE.md or an AGENTS.md, you do not need to rewrite the instructions for each tool. Codex reads AGENTS.md ^[7]. Claude Code can read AGENTS.md in place of CLAUDE.md when the latter is missing ^[1]. TRAE reads both ^[18].

Limits and common questions

• Prices are in different currencies and units. Codex and Claude Code prices are in US dollars. Quotas are counted differently: some by messages, some by dollar usage or tokens. Compare the unit, not just the number ^[9][14][19].
• Free access may be narrow. The Gemini CLI README says there is no free tier. TRAE's Free plan can run out within a month ^[10][20].
• Model lists change. Each tool changes its default and its model list over time, so check the settings page before you rely on a specific model ^[4][8][17].

Next steps

Start with the official quick start of the tool you choose. The Claude Code quick start is the place to begin ^[2]. The Codex CLI README covers installation and sign-in ^[6]. The Gemini CLI installation guide covers Node.js and the commands ^[11]. The TRAE get-started guide covers installation and platform support ^[16].