Codex vs Claude Code (November 2026): Pricing, Limits and Benchmarks

OpenAI Codex and Claude Code are now the two biggest AI coding agents. Both read your repo, edit files, run tests and open pull requests.

They differ on price, limits, models and how they read project instructions. This guide compares them as of November 7, 2026, using official pricing pages and docs.

You will also see what changed on October 29 for Codex, which benchmarks matter, and a clear pick for each kind of developer.

Codex vs Claude Code at a glance

Both tools are agents, not autocomplete. You describe a task in plain English, and the agent plans, edits, runs commands and shows you a diff.

The table below sums up the main differences. Each row is explained in its own section further down.

Pricing and plans (November 2026)

Neither agent is sold on its own. Codex comes with ChatGPT plans, and Claude Code comes with paid Claude plans. Both can also run on a pay-per-token API key.

Codex plans

• Free ($0) and Go ($8/month): GPT-6 Luna in the desktop app only, subject to rollout.
• Plus ($20/month): Codex on web, CLI, IDE and iOS, plus automatic code review and Slack. You can buy extra credits.
• Pro ($100, $200 or $500/month): no 5-hour limit, only a weekly one. Ultrafast mode is on the $500 plan.
• Business: $20 per user per month billed yearly, or $25 monthly, for 2 or more users.
• Enterprise and Edu: contact sales. Flexible pricing plans have no fixed rate limits.

Claude Code plans

• Free: Claude Code is not included.
• Pro ($20/month, or $17/month yearly): the base allowance.
• Max ($100/month): 5x Pro usage every 5 hours and every week.
• Ultra ($200/month): 10x Pro weekly, and no 5-hour limit for verified accounts.
• Team: Standard seats $25 monthly or $20 yearly. Premium seats $125 or $100, with 5x Standard usage.
• Enterprise: $20 per seat per month billed yearly, plus usage at API rates.

Claude Code usage shares one pool with Claude chat. My Claude Code pricing guide has worked cost examples, and the Codex guide covers every Codex surface and credit.

Key point: at $20 the two are close. Above that, Codex has three Pro tiers up to $500, while Claude tops out at $200 for individuals.

The October 29 Codex limit changes

On October 29, 2026, Codex lead Thibault Sottiaux announced bigger allowances on OpenAI’s paid tiers. Prices did not change, and no renewal was needed.

• Pro 100 and Pro 500: twice their previous Work and Codex allowance.
• Pro 200: a cut announced in a September 29 email was cancelled. Accounts on the lower allowance were raised to the higher one.
• Business Premium: also raised.
• Plus: unchanged.

OpenAI still does not publish fixed message counts for Pro. For Plus, its pricing page gives estimated ranges of local messages per 5 hours, such as 15 to 160 for GPT-6.1 Sol and 5 to 45 for GPT-6 Astra.

Claude had its own change on November 1. The old Max 20x plan was renamed Ultra, and verified Ultra accounts lost the 5-hour window.

Models and API prices

Each agent runs its own company’s models. You can switch models inside a session to trade speed and cost for quality.

Recent model news on the Codex side:

• GPT-6.1 Sol: launched at DevDay 2026. OpenAI says it nears Astra on several evals at one-fifth of Astra’s token price.
• GPT-6.1 Terra: added in the week of October 26 for Plus, Pro, Business, Enterprise and Edu.
• GPT-5.5: retired from Codex with ChatGPT sign-in on October 14, 2026. It stays on the API.

On the Claude side, Opus 5.5 is the main coding model and Sonnet 5.5 is the cheaper option. Fable 5.1 is on Max, Ultra and Team Premium, within a share of the weekly limit.

CLI, IDE, desktop and cloud

Both tools now run almost everywhere. The difference is in the details of each surface.

Codex surfaces

• CLI: open source on GitHub under Apache-2.0. Commands include codex exec for scripts, codex cloud for cloud tasks and codex mcp for MCP servers.
• Desktop app: part of the ChatGPT desktop app, and the only surface on Free and Go.
• IDE extension: works in VS Code and other editors, with the same account and limits as the CLI.
• Cloud: runs tasks in OpenAI’s sandbox. DevDay added reusable cloud environments you set up once.
• Code review: tag Codex on a GitHub pull request, or turn on automatic reviews.

Claude Code surfaces

• CLI: the full-featured terminal tool. Version 2.1.319 is the newest in the official changelog as of November 7.
• IDE: extensions for VS Code, Cursor and JetBrains, with inline diffs and plan review.
• Desktop app: runs several sessions side by side and schedules recurring tasks.
• Web: cloud sessions on Pro, Max, Ultra and Team. See my Claude Code on the web guide.
• Code review: the managed Code Review service is for Team and Enterprise. The local /code-review command works on any plan.

Note: Claude Code can also run through Amazon Bedrock, Google Cloud and Microsoft Foundry. That matters if your company already buys AI through one of those clouds.

AGENTS.md vs CLAUDE.md

Both agents read a markdown file of project rules at the start of each session. The file names and loading rules differ, which trips up teams that use both.

How Codex loads AGENTS.md

• Global file: ~/.codex/AGENTS.md, or AGENTS.override.md if present.
• Project files: Codex walks from the Git root down to your current folder and takes at most one file per folder.
• Order: files closer to your current folder come later and win.
• Size cap: 32 KiB combined by default, set by project_doc_max_bytes.

How Claude Code loads CLAUDE.md

• Project file: ./CLAUDE.md or ./.claude/CLAUDE.md, plus a private CLAUDE.local.md.
• AGENTS.md support: since v2.1.277, Claude Code reads AGENTS.md, but only when no CLAUDE.md exists in your folder or above it.
• Share one file: put @AGENTS.md inside a small CLAUDE.md to import it.

Tip: if your team uses both agents, write your rules once in AGENTS.md. Then add a one-line CLAUDE.md that imports it. My AGENTS.md guide has a full example.

Benchmarks with sources

Benchmarks test models, not the agent apps around them. Still, they show where each model is strong. The numbers below come from Vals AI’s head-to-head page for Claude Opus 5.5 and GPT-6.1 Sol.

Opus 5.5 wins most coding tests on this page. GPT-6.1 Sol wins SRE Bench clearly, and it is far cheaper and faster on almost every run.

OpenAI’s own launch numbers tell a different story. On AutomationBench, it reports GPT-6.1 Sol at 35.4% at medium effort, about 2.2 points above Opus 5.5, at roughly a third of the cost per task.

Warning: vendor benchmarks pick tests that flatter their own model. Treat them as hints, and test both agents on your own repo. The November 2026 AI model leaderboard tracks more scores.

How to try both on one repo

The best test is a real task from your own backlog. Here is a fair way to compare them in about an hour.

1. Pick one task. Choose a small, real issue with tests, such as a bug fix or a new endpoint.
2. Install Codex. Run npm install -g @openai/codex, or brew install --cask codex on a Mac.
3. Install Claude Code. On macOS or Linux, run curl -fsSL https://claude.ai/install.sh | bash. Then check it with claude --version.
4. Share instructions. Put your rules in AGENTS.md, and add a CLAUDE.md that contains @AGENTS.md.
5. Make two branches. Create one branch for each agent, so the results never mix.
6. Run the same prompt. Start codex on one branch and claude on the other, and paste the same task.
7. Compare the results. Check whether tests pass, how many changes you had to fix, and how much of your limit each used.

In Codex, /status shows your remaining limits. In Claude Code, /usage shows yours.

Which one should you pick?

There is no single winner. Pick based on budget, how you work and what your team already pays for.

• You want to start free: Codex. Its Free plan includes the desktop app with GPT-6 Luna.
• You already pay for ChatGPT Plus: try Codex first, since it costs you nothing extra.
• You want the strongest coding scores: Claude Code with Opus 5.5 leads most coding tests on Vals AI.
• You care most about cost per task: Codex with GPT-6.1 Sol is much cheaper per run on the same tests.
• You run agents all day: compare Claude Ultra at $200 with Codex Pro 200. Both drop the 5-hour window.
• You want open source tooling: the Codex CLI is Apache-2.0, so you can read and fork it.
• Your company uses AWS or Google Cloud: Claude Code runs through Bedrock and Google Cloud.

Many developers keep both on the $20 tiers. Then they use whichever has limit left, and save the harder tasks for the agent that handles them best.

Troubleshooting

Codex CLI will not let me sign in on the Free plan

On Free and Go, Codex works only in the desktop app. Upgrade to Plus, or sign in to the CLI with an OpenAI API key and pay per token.

Claude Code says my plan does not include it

The Claude Free plan does not include Claude Code. Upgrade to Pro, or sign in with a Claude Console account that has API credits.

Claude Code ignores my AGENTS.md

A CLAUDE.md somewhere in your folder or above it takes priority. Add @AGENTS.md to that CLAUDE.md, and update to v2.1.277 or later.

Codex seems to forget part of my AGENTS.md

Codex stops adding instruction files at 32 KiB in total by default. Shorten the files, split rules into subfolders, or raise project_doc_max_bytes in ~/.codex/config.toml.

My Pro limit did not change after October 29

The doubling applied to Pro 100, Pro 500 and Business Premium. On Pro 200, only the lower allowance was raised, so long-time subscribers may see no change.

Frequently Asked Questions

Is Codex or Claude Code better for coding?

On Vals AI’s tests, Opus 5.5 beats GPT-6.1 Sol on most coding benchmarks, including Terminal-Bench 4.0. GPT-6.1 Sol is much cheaper per run, so the better pick depends on your budget.

Is Codex free to use?

Yes, in a limited way. The ChatGPT Free and Go plans include the Codex desktop app with GPT-6 Luna, subject to rollout. The CLI, IDE and web need Plus or an API key.

How much does Claude Code cost?

Claude Code starts with Pro at $20 a month, or $17 billed yearly. Max is $100 and Ultra is $200, or you can pay per token with an API key.

What changed for Codex on October 29, 2026?

OpenAI doubled Work and Codex allowance on Pro 100, Pro 500 and Business Premium. It also cancelled a planned cut to Pro 200, while Plus stayed the same.

Does Claude Code read AGENTS.md?

Yes, since v2.1.277, but only when no CLAUDE.md exists in your folder or above it. To use both, import AGENTS.md from a short CLAUDE.md.

Can I use Codex and Claude Code in the same project?

Yes. Keep shared rules in AGENTS.md, import it from CLAUDE.md, and give each agent its own branch so their changes do not clash.

Is the Codex CLI open source?

Yes. The Codex CLI repository on GitHub uses the Apache-2.0 license. Claude Code is not open source.

Sources

• OpenAI: Codex pricing
• OpenAI: What’s new in ChatGPT and Codex
• OpenAI: Custom instructions with AGENTS.md
• GitHub: openai/codex
• Anthropic: Claude pricing
• Claude Code docs: CLAUDE.md and AGENTS.md
• Claude Code docs: Overview
• Vals AI: Claude Opus 5.5 vs GPT-6.1 Sol
• Data Studios: OpenAI doubles Pro and Business Premium limits
• InfoQ: OpenAI DevDay 2026