Claude Code vs Codex (2026): Differences and Pricing

The short answer: Claude Code is Anthropic’s coding agent and Codex is OpenAI’s. Both run in the terminal, in your IDE, in a desktop app and in the cloud, and both connect to tools over MCP. Claude Code runs Anthropic’s models and comes with Claude Pro, Max, Team and Enterprise. Codex runs OpenAI’s models and comes with ChatGPT plans, Free included. Many teams use both.

At a glance

Claude Code Codex
Made by Anthropic OpenAI
Runs in Terminal, VS Code and Cursor, JetBrains, desktop app, web, Slack, iOS and Android CLI, IDE extension (VS Code, Cursor, Windsurf), Xcode and JetBrains integrations, ChatGPT desktop, web and mobile, cloud
Plans Claude Pro ($20/month, or $17/month billed annually), Max, Team and Enterprise; not the Free plan Included in ChatGPT Free, Go, Plus, Pro, Business, Edu and Enterprise
Models Anthropic’s: Opus, Sonnet, Haiku and Fable OpenAI’s GPT models, switched with /model
MCP Yes: HTTP (recommended), stdio and WebSocket servers Yes: stdio and Streamable HTTP servers, set in config.toml
Instruction file CLAUDE.md; reads AGENTS.md when there’s no CLAUDE.md AGENTS.md
Sandboxing OS-level sandbox for Bash (/sandbox); permission modes; cloud sessions in an isolated VM OS-enforced sandbox: read-only, workspace-write or full access; network off by default
Open source Not stated The Codex CLI is (openai/codex); the IDE extension and cloud aren’t
Parallel work Subagents, background agents and agent teams (experimental) Many agents in parallel; cloud tasks

Where they differ

Plans and price. Claude Code needs a paid Claude plan: Pro is $20 a month, or $17 a month billed annually, and Max and Team premium seats raise the limits. Codex comes with every ChatGPT plan, from Free and Go ($8 a month) up, and OpenAI says ChatGPT Work and Codex “share pricing, credits, and usage limits.” Both can also be paid for by API usage instead.

Instructions. Codex “reads AGENTS.md files before doing any work.” Claude Code reads CLAUDE.md, and by default reads AGENTS.md only when there’s no CLAUDE.md. A shared AGENTS.md, referenced from CLAUDE.md, gives both agents the same rules.

Safety controls. Claude Code combines permission modes (from read-only to auto, where a classifier reviews actions) with an OS-level sandbox for Bash. Codex pairs an OS-enforced sandbox, with network access off by default, with approval policies.

Openness. OpenAI publishes the Codex CLI as open source. Anthropic doesn’t say the same of Claude Code.

Which is better for what

Pick by the plan your team already pays for and the model family you trust for your codebase. Teams that already pay for Claude get Claude Code included; teams on ChatGPT get Codex. Both handle bug fixes, refactors, tests and multi-file features, and both can run work in the cloud while you’re away.

You don’t have to pick one

Plenty of teams run both: one engineer in Claude Code, another in Codex, on the same repo. The hard part is keeping them from colliding. There are four ways to handle it, from git worktrees to shared threads, in how to run Claude Code and Codex together.

elvex is the shared-thread option. Each person connects their own agent over MCP, so Claude Code and Codex read the same thread, claim and hand off to-dos, warn each other before a change lands, and bring a person in only when there’s a decision to make. Set-up for both is on the elvex MCP page. For every option, see how to make Claude Code and Codex talk to each other.

Frequently Asked Questions

Is Claude Code free?

No. Claude Code isn’t on the Claude Free plan. It comes with Pro ($20 a month, or $17 a month billed annually), Max, Team and Enterprise, or pay-as-you-go through a Claude Console account at API prices.

Is Codex free?

Codex is included in ChatGPT plans, Free and Go among them, with higher limits on Plus, Pro and Business.

Do Claude Code and Codex use the same instructions file?

Codex reads AGENTS.md. Claude Code reads CLAUDE.md, and reads AGENTS.md when there’s no CLAUDE.md. Keep shared rules in AGENTS.md and point CLAUDE.md at it.

Do both support MCP?

Yes. Claude Code connects to HTTP, stdio and WebSocket MCP servers; Codex to stdio and Streamable HTTP servers. Either can connect to a remote server such as elvex.

Can Claude Code and Codex work on the same project?

Yes. Give each its own git worktree and branch, and give them a way to see each other’s plans, such as a shared elvex thread.

Sources (checked Nov 20, 2026): Claude Code overview (Anthropic); Claude pricing (Anthropic); Claude Code memory: CLAUDE.md and AGENTS.md (Anthropic); Claude Code security (Anthropic); Codex (OpenAI); Codex pricing (OpenAI); Codex IDE extension (OpenAI); Codex AGENTS.md (OpenAI); Codex sandboxing (OpenAI); Codex open source (OpenAI). Product details change; check each vendor's page for the latest.

Bring your agent. Bring your team.