Claude Code and Codex are the two coding agents most developers end up choosing between. Both run in your terminal, both edit your repo, and both start at $20 a month.
Most comparisons that rank for this were written in spring and summer. Since then Claude Code’s default model became Opus 5.5, OpenAI released GPT-6.1 Sol, and ChatGPT’s plans now run from $8 to $500. This guide uses the vendors’ own pages, read on October 16, 2026.
We make Context Board, a live board that shows what’s left of both plans side by side, so we use both every day. That’s the bias, and it’s also where the last section comes from: plenty of people stop choosing and pay for both.
- Price: both start at $20. Claude Code comes with Claude Pro, Max 5x ($100) and Max 20x ($200). Codex comes with every ChatGPT plan, from Go at $8 to Pro at $100, $200 or $500.
- Pick Claude Code for long, careful work in one repo. Opus 5.5 by default, plan mode, hooks, skills and subagents.
- Pick Codex for an open-source CLI, a cheaper way in, and cloud tasks that run on your GitHub repos while your laptop sleeps.
- Limits decide most days: both use five-hour and weekly windows. Check with
/usagein Claude Code and/statusin Codex. - Using both? Context Board shows both plans’ limits and every session waiting on you on one screen, for $6 a month at the launch price.
Claude Code vs Codex at a glance
| Claude Code | Codex | |
|---|---|---|
| Made by | Anthropic | OpenAI |
| Cheapest plan with it | Claude Pro, $20/mo | ChatGPT Go, $8/mo (desktop app, rolling out) |
| Top plan | Max 20x, $200/mo | Pro, up to $500/mo |
| Default model | Opus 5.5 | A recommended model, GPT-6.1 Sol today |
| Limits | Five-hour and weekly, shared with Claude chat | Five-hour and weekly on Plus; Pro has no five-hour limit for now |
| Check usage | /usage, or claude.ai/settings/usage | /status, or chatgpt.com/codex/settings/usage |
| Open source | Public repo, all rights reserved | CLI is Apache-2.0 |
| Instructions file | CLAUDE.md, and it can read AGENTS.md | AGENTS.md |
| Runs in | Terminal, VS Code, JetBrains, desktop app, web, mobile, Slack, GitHub Actions | Terminal, VS Code-family IDEs, desktop app, ChatGPT web and mobile, Codex Cloud, Slack |
| MCP | Yes | Yes |
Price and plans
Both tools come inside a chat subscription rather than as their own product. So the real question is which subscription you’d rather pay for.
Claude Code is in every paid Claude plan, per Claude’s pricing page:
- Pro: $20 a month, or $17 a month billed yearly.
- Max 5x: $100 a month, five times Pro’s usage.
- Max 20x: $200 a month, twenty times Pro’s usage, per Anthropic’s Max plan article.
- Team: $25 a seat monthly for Standard, $125 for Premium with five times the usage.
Codex is in every ChatGPT plan, per OpenAI’s Codex pricing docs:
- Free and Go ($8 a month): GPT-6 Luna in the Codex desktop app, which OpenAI says is still rolling out.
- Plus: $20 a month, with GPT-6.1 Sol and GPT-6 Luna.
- Pro: from $100 a month, with tiers at $100, $200 and $500. The $500 tier adds Ultrafast mode.
- Business: $25 a user monthly, or $20 billed yearly.
At $20 they’re level. Below it, Codex has a way in that Claude Code doesn’t. Above it, Claude’s $200 Max 20x and ChatGPT’s $200 Pro are the usual pick for people who code all day.
Both let you keep going past a limit for extra money. Claude calls them usage credits, and Codex sells credits to Plus and Pro.
Usage limits: how they work and where to check
Limits are where most people actually feel the difference. You don’t notice the model until you notice it stopped.
Claude Code on Pro and Max has two windows. A five-hour session limit, and a weekly limit, both shared with Claude chat, per Anthropic’s support article. Max also has a separate weekly limit for Fable. Type /usage in Claude Code to see the bars, or open claude.ai/settings/usage. Our guide to checking Claude Code usage covers every way in.
Codex on Plus and standard Business counts local messages per five-hour window, and weekly limits can apply too. OpenAI publishes rough ranges rather than fixed caps. On Plus, GPT-6.1 Sol allows roughly 15 to 160 local messages per five hours, and the lighter GPT-6 Luna 350 to 3,000. OpenAI also says Pro plans currently have no five-hour limit. Type /status in a Codex session, or open chatgpt.com/codex/settings/usage. The Codex usage guide has the details.
Neither company publishes a fixed number of prompts. How far a window goes depends on the model, the size of your repo and how much the agent reads.
Models and benchmarks
Claude Code’s default resolves to Opus 5.5 on Pro, Max, Team and Enterprise, per its model docs. You can switch to Fable 5.1 for the hardest jobs, Sonnet 5.5 for speed, or Haiku 5.5 for small edits.
Codex uses a recommended model rather than naming a fixed default. Right now that’s GPT-6.1 Sol, which OpenAI pitches as close to GPT-6 Astra for less. GPT-6 Luna is the fast, cheap option for high-volume work.
On benchmarks, be careful. Anthropic’s Opus 5.5 announcement reports 66.4% on Terminal-Bench 4.0, against 57.9% for GPT-6 Astra. That table predates GPT-6.1 Sol, and each lab runs its own harness. A benchmark tells you the models are close. Your repo tells you which one suits you.
Where each one runs
Both started as terminal tools and now run almost everywhere.
Claude Code runs in the terminal, VS Code and Cursor, JetBrains, the Claude desktop app, on the web at claude.ai/code and on mobile. It also runs in Slack and GitHub Actions, per the Claude Code docs.
Codex runs in the terminal, IDE extensions for VS Code, Cursor and Windsurf, its own desktop app, ChatGPT on the web and phone, and Slack.
The big difference is Codex Cloud. You point a task at a GitHub repo, it runs in its own isolated workspace while your computer sleeps, and you review the result and open a PR. Claude Code has web sessions and GitHub Actions for similar work, but Codex Cloud is the more direct “hand it off and walk away” flow.
Configuration, MCP and extending them
Instructions: Claude Code reads CLAUDE.md, and its docs say it can read a repo’s AGENTS.md in its place. Codex uses AGENTS.md, and /init writes one for you. So one AGENTS.md can now serve both.
MCP: both connect to MCP servers, so the same tools (your database, GitHub, a board) work in either. Add one with claude mcp or codex mcp.
Extending: Claude Code has the deeper kit. Hooks run your own scripts before or after a tool call, skills package instructions, subagents split work, and plan mode makes it propose before it touches anything. Codex has skills and can delegate to specialised agents too, but Claude Code’s hooks and plan mode are the more mature pieces.
Open source: the Codex CLI is open source under Apache-2.0, so you can read and fork it. Claude Code’s repo is public, but its license says all rights reserved.
Safety: sandboxes and approvals
Codex is strict by default. Network access is off, a Git folder starts in a mode that can edit the workspace but asks before anything else, and a folder outside Git starts read-only. You can choose read-only, workspace-write or full access, and when it asks for approval.
Claude Code works from permission modes. It asks before edits and commands unless you allow them, and plan mode keeps it read-only until you approve a plan.
Both are safe enough to start with. The difference is in the defaults. Codex locks down first, and Claude Code asks first.
Which should you pick?
- You want one tool and code most days: start on whichever chat plan you already pay for. Both are good, and $20 gets you far enough to find out.
- You do long refactors in one big repo: Claude Code. Opus 5.5, plan mode and hooks reward careful, multi-step work.
- You want to hand off tasks and come back later: Codex. Cloud tasks on GitHub are the cleanest way to do it.
- You care about open source or want a cheaper start: Codex, from $8 on Go.
- You keep hitting limits: use both. Two $20 plans often go further than one, as long as you can see which has room.
Using Claude Code and Codex together
Plenty of developers stop choosing. They keep Claude Code for the hard work and give Codex the batch jobs, or the other way round. Both read AGENTS.md and both speak MCP, so the repo doesn’t care which agent touched it last.
What breaks is keeping track. Two plans, each with a five-hour window and a week. Sessions running in three terminals, one of them waiting for an OK you haven’t seen. That’s the job Context Board does.
Context Board is ours, so weigh the bias. It’s a live board for the business, and its AI usage widget shows what’s left of each plan’s five-hour window and week, Claude Code and Codex side by side. The rings are green at half or more, amber from 20% and red under 20%, each with when it resets. Needs you lists every session waiting for an answer, an OK or a plan, across every computer you link.
- Codex · docsasks to run the test suite3m
Claude Code is at 18% for this window. Test fixes go to Codex until 4 PM.
A small runner on your computer reads what each app last heard from its own servers. You set it up once with one line from Connect. Only names, states and percentages leave your computer, never what was said in a session. The Agents template has AI usage, Needs you, Git activity and Agent activity on it already.
The board’s chat runs on your own Claude Code or Codex plan too, so asking it to rearrange the board or add a widget spends the plan you already pay for.
If you already track Claude Code with ccusage or Grafana, the Claude Code dashboard guide compares them, and Context Board vs ccusage goes line by line. The Claude Code usage page shows the setup.
Try Context Board free for 3 days, then $6 a month at the launch price, and put both plans on one screen.