Claude Code vs Codex: which coding agent suits you?

Prices in Germany (as of October 2026)

Claude Code: according to Anthropic, Claude Code is included in all paid Claude plans but not in the free plan. According to the German pricing page, Pro costs 20 dollars a month or 200 dollars paid annually in advance, which works out at 17 dollars a month, and Max starts at 100 dollars with five or twenty times the usage of Pro. Anthropic shows dollar prices excluding any taxes and says prices vary by region, with tax sometimes added only at checkout. In July 2026 Notebookcheck put Claude Pro at around 18 euros and Max from around 90 euros; that is not a current German final price. The amount shown at checkout is what counts.

Codex: according to OpenAI, Codex is included in all ChatGPT plans, including Free and Go, but only with limited access there. The German pricing page lists Go at 8 euros, Plus at 23 euros and Pro from 103 euros a month, with Pro offering three usage levels to choose from. Notebookcheck still listed 229 euros for Pro in July, so the levels are new. Tasks in the Codex cloud start with Plus, according to OpenAI for eligible accounts and depending on the gradual rollout and settings.

For teams: according to Anthropic, Claude Team costs 20 dollars per seat and month with annual billing or 25 dollars monthly. According to OpenAI, ChatGPT Business is available from two users, and Business and Enterprise customers can buy extra Codex credits.

Where the agents work

A coding agent is an AI program that reads files, writes code and runs commands by itself. Our article on coding agents for non-programmers explains what this means in general. Here the question is how to choose between the two best-known ones.

According to Anthropic, Claude Code runs in the terminal, the text window for commands, in development environments such as VS Code and JetBrains, in the Claude desktop app and in the browser. The program runs locally but sends all prompts and outputs to Anthropic in encrypted form. Cloud sessions run in Anthropic’s virtual machines, into which your code repository is copied.

According to OpenAI, Codex is available as a command-line program, as an extension for VS Code and similar editors, in the Codex mode of the ChatGPT desktop app and on the web. Local workflows run on your device, cloud tasks on OpenAI’s computers. According to OpenAI, these keep running even while your computer is asleep.

Usage limits

According to Anthropic, Claude Code shares its limits with your other Claude chats. They reset in a five-hour window, and paid plans add weekly limits. Anthropic does not give a fixed number of messages. If you need more, you can switch on usage credits at API prices.

According to OpenAI, Codex shares one allowance with ChatGPT Work and other agent features. How quickly it runs out depends on the model, the task and whether it runs locally or in the cloud. According to OpenAI, Plus and Pro customers can buy credits without changing plans.

Some Codex extras are not available in Europe according to OpenAI: the website feature Sites is not available in the European Economic Area, Switzerland and the UK, and Record & Replay, which saves a demonstrated workflow as a reusable skill, initially excludes the EU.

Privacy: what happens to your code

Anthropic: on Free, Pro and Max, Anthropic says you decide whether your data may be used to improve future models, including data from Claude Code. If you allow it, Anthropic keeps the data for five years, otherwise for 30 days. On Team, Enterprise and the API, Anthropic says it does not train on code or prompts. Claude Code also stores session transcripts locally on your computer, for 30 days by default; you can change this period. Sessions you started or last continued in the Claude desktop app or in Cowork are exempt from this limit by default.

OpenAI: on Plus and Pro, OpenAI says conversations may be used to improve models unless you switch off training in Data controls. This setting also applies to Codex tasks. Additional context from your Codex environments, however, is controlled by a separate switch called Include environments in the Codex settings, which switching off training in ChatGPT does not change. On Business, Enterprise and the API, OpenAI says it does not use inputs and outputs for training by default. According to the pricing page, data residency in a specific region such as the EU is not included in the personal plans.

Regardless of the provider, the BSI and the French agency ANSSI warn of two risks of AI coding assistants: the confidentiality of what you enter and security flaws in the generated code. So check what the agent changes before you accept it.

Which agent for whom?

The following is an editorial recommendation based on the providers’ information, not our own performance test.

Codex suits you if you already use ChatGPT, want to try it for free first or want to hand tasks to the cloud that keep running while your computer is off. With Plus at 23 euros you get chat and agent in one subscription.

Claude Code suits you if you already work with Claude or want an agent that works the same way in the terminal, your development environment and the desktop app. In a Claude subscription, Claude Code is included from Pro; there is no free option here. Alternatively, according to Anthropic, you can pay as you go with an Anthropic Console account, that is via the API, the access route for programs that is billed by actual usage.

For work projects with other people’s code, the Team or Business plans are the safer choice with both, because according to the providers they do not train on your data by default.