Can I use my own models or third-party providers with either tool?

No — both are single-vendor. Codex CLI uses OpenAI models only (GPT-5.x / o-series). Claude Code uses Anthropic Claude models only (Opus 4.8 / Sonnet / Haiku). Neither supports swapping in a different provider's model.

How does MCP configuration differ between the two?

Codex CLI configures MCP servers in ~/.codex/config.toml using STDIO or streaming HTTP transport, and Codex itself can act as an MCP server for other tools. Claude Code treats MCP as a native layer with connectors, and stacks on top of that with subagents, hooks, skills, and plugins for more complex orchestration.

What are Codex Cloud tasks, and how do they compare to Claude Code subagents?

Codex Cloud tasks let you offload work to run outside your local terminal session — confirm current scope and per-plan availability on the OpenAI developer site. Claude Code's subagents and agent teams allow you to spin up coordinated multi-agent workflows where one agent can orchestrate others on multi-step engineering tasks. They serve overlapping use cases but with different architecture.

Is Claude Code free after June 22, 2026?

Claude Code was included at no extra cost on Claude Pro/Max/Team plans until June 22, 2026, after which it moves to a usage-credits model. Confirm the exact credits allocation and pricing on Anthropic's site, as this is actively changing.

How do I give each tool project-level instructions?

Codex CLI reads an AGENTS.md file in your repository root for workspace-specific instructions. Claude Code supports a project-level instructions file at the repo root as well — confirm the exact filename on Anthropic's site. Both tools follow a markdown-file-in-the-repo-root convention, but the filenames and schemas differ between the two.

Comparison

OpenAI Codex CLI vs Claude Code

Codex CLI and Claude Code are the flagship terminal coding agents from OpenAI and Anthropic respectively, and their deepest difference is not feature parity — it is which model ecosystem you are betting on. Codex CLI ships an open-source Rust client and leans on Codex Cloud tasks for async offloading; Claude Code is a proprietary agent with a richer subagent and orchestration layer. Your existing ChatGPT or Claude plan subscription is often the deciding factor before any technical comparison matters.

By DK, Editor · Last verified: 2026-06-20 · How we test · No hands-on score yet — comparison is on documented facts

At a glance

Dimension	Codex CLI	Claude Code
Form factor	Terminal CLI + Codex Cloud tasks	Terminal agent + IDE + desktop app
Open source	CLI is open source (Rust); service is OpenAI	Proprietary (Anthropic)
Free tier	Included with ChatGPT Plus/Pro/Team/Enterprise allotments; API usage-priced (confirm current)	Included on Claude Pro/Max/Team plans until 2026-06-22, then moves to usage credits (confirm current)
Models / provider	OpenAI only (GPT-5.x / o-series family)	Anthropic Claude only (Opus 4.8 / Sonnet / Haiku)
MCP support	Yes — configure servers in ~/.codex/config.toml (STDIO or streaming HTTP); Codex can also run as an MCP server	Yes — MCP-native via connectors; layered with subagents, hooks, skills, and plugins
Cloud / background work	Codex Cloud tasks (confirm scope and per-plan availability on the vendor's site)	Subagents and agent teams (orchestrated multi-agent workflows)
Workspace instructions	AGENTS.md file in the repo root	Project-level instructions file in the repo root (confirm exact filename on the vendor's site)
Status	Active (June 2026: improved startup, OAuth MCP refresh)	Active
Best for	Developers in the OpenAI ecosystem wanting terminal + cloud tasks with MCP	Developers in the Anthropic ecosystem wanting powerful multi-step agentic work with subagents + MCP

Facts compiled from public sources and verified 2026-06-20 — pricing and models change often, so confirm current details on each vendor’s site. No hands-on score is shown; this is a documented-fact comparison.

Verdict

If your team already pays for ChatGPT Plus, Pro, or Team, Codex CLI's allotment-based pricing means you can start without a separate budget line — but you are locked to OpenAI models and Codex Cloud for background work. Claude Code's subagent and agent-team layer is more articulated for complex multi-step engineering workflows, and its MCP integration is deeply native across connectors, hooks, and skills; the tradeoff is single-vendor model lock-in to Anthropic's Claude family and a pricing shift to usage credits after June 22, 2026. Codex CLI's open-source Rust client is a meaningful differentiator if your security or audit process requires inspectable client code. Both tools require you to commit to one lab's model ecosystem, so the right choice hinges on which models you already use and trust for code, not on feature parity alone.

Choose Codex CLI if you are already on a ChatGPT Plus/Pro/Team/Enterprise plan and want terminal + cloud task offloading without adding a new subscription, and you prefer OpenAI's GPT-5.x or o-series models for code generation.
Choose Claude Code if you want the most fully-featured agentic layer — subagents, agent teams, hooks, skills, and plugins — and are comfortable with Anthropic's Claude models (Opus 4.8 / Sonnet / Haiku) as your sole option.
Choose Codex CLI if open-source CLI code matters to your security or audit requirements — the Rust CLI is public; Claude Code's client is proprietary.
Choose Claude Code if your workflow benefits from IDE and desktop app form factors in addition to the terminal, rather than a terminal-first tool with cloud task offloading.

Our AI agents

From the team behind AI Coding Hub — agents that pick up where the code ends:

FAQ

Can I use my own models or third-party providers with either tool?: No — both are single-vendor. Codex CLI uses OpenAI models only (GPT-5.x / o-series). Claude Code uses Anthropic Claude models only (Opus 4.8 / Sonnet / Haiku). Neither supports swapping in a different provider's model.
How does MCP configuration differ between the two?: Codex CLI configures MCP servers in ~/.codex/config.toml using STDIO or streaming HTTP transport, and Codex itself can act as an MCP server for other tools. Claude Code treats MCP as a native layer with connectors, and stacks on top of that with subagents, hooks, skills, and plugins for more complex orchestration.
What are Codex Cloud tasks, and how do they compare to Claude Code subagents?: Codex Cloud tasks let you offload work to run outside your local terminal session — confirm current scope and per-plan availability on the OpenAI developer site. Claude Code's subagents and agent teams allow you to spin up coordinated multi-agent workflows where one agent can orchestrate others on multi-step engineering tasks. They serve overlapping use cases but with different architecture.
Is Claude Code free after June 22, 2026?: Claude Code was included at no extra cost on Claude Pro/Max/Team plans until June 22, 2026, after which it moves to a usage-credits model. Confirm the exact credits allocation and pricing on Anthropic's site, as this is actively changing.
How do I give each tool project-level instructions?: Codex CLI reads an AGENTS.md file in your repository root for workspace-specific instructions. Claude Code supports a project-level instructions file at the repo root as well — confirm the exact filename on Anthropic's site. Both tools follow a markdown-file-in-the-repo-root convention, but the filenames and schemas differ between the two.