Quick answer

Three terminal coding agents from the three largest labs. Claude Code (Anthropic) is the most mature, on the strongest coding models, with hooks, skills, MCP, and Windows support. Codex CLI (OpenAI) gained voice steering, an /agents view for delegating several tasks, cloud environments, and automatic code review at DevDay 2026, and is bundled with ChatGPT plans. Muse Code (Meta) launched in beta on August 5, 2026 on Muse Spark 1.2 with persistent background subagents in isolated git worktrees and an append-only event log that resumes after a crash, on macOS and Linux; coverage reports it left beta on August 31 with plans from $5 a month. Google's equivalent is now the Antigravity CLI.

A year ago "terminal agent" meant Claude Code and a handful of open-source tools. Now every lab has one, and the differences are in architecture and price rather than the basic loop of read, plan, edit, test.

Models

  • Claude Code: Anthropic models only; Sonnet 5.5 and Opus 5.5 are the current defaults for most users
  • Codex CLI: OpenAI models; GPT-6.1 Sol and GPT-6 Astra, with Ultrafast mode on Pro 500 and Enterprise
  • Muse Code: Meta's Muse models only, Muse Spark 1.2 at launch
  • None of the three is model-agnostic; for that you want Cline, Roo Code, Aider, or Hermes

Architecture

Muse Code's distinctive idea is that background subagents stay alive for the whole session and accumulate context, instead of restarting per task, and that large jobs split into parallel subagents in isolated git worktrees. Its local event log records every model call, tool run, approval, and edit, so a crash resumes exactly where it stopped. Claude Code has subagents, hooks, and skills with the deepest MCP integration. Codex CLI added an /agents view and session resume at DevDay, and pairs with Codex in Cloud for remote environments.

Platforms and price

  • Claude Code: macOS, Linux, Windows; usage through Claude Pro and Max plans or API billing
  • Codex CLI: macOS, Linux, Windows; included with ChatGPT Plus, Pro, Business, Enterprise, Edu
  • Muse Code: macOS and Linux; monthly plans reported from $5, with an SDK in developer preview

Which to use

  • Default: Claude Code, for the model, the ecosystem, and the track record
  • Already on ChatGPT: Codex CLI, since it is included and integrates with Codex in Cloud and code review
  • Large monorepos, auditability, or price: try Muse Code, with the caveat that several claims rest on launch-week coverage
  • On Gemini: the Antigravity CLI, which replaced Gemini CLI

Related reading

Google Antigravity review Cursor vs Windsurf vs Claude Code OpenAI DevDay 2026: Codex updates

Bottom line

Claude Code remains the one to beat. Codex CLI is the free upgrade for ChatGPT subscribers. Muse Code is the most interesting architecture of the three and the least proven. Run the same task through two of them; the terminal makes that cheap.