Coding with AI: ChatGPT vs Claude vs Copilot as a programming assistant

Programming is the task where AI subscriptions differ the most. For writing and summarizing, the major models are hardly apart, but with code the difference is not just in the model: it is in where you use it. A chat window you copy code into and out of is something different from an assistant that sees your whole project in your editor, and that in turn is different from an agent that completes a task on its own.

This comparison goes through the five options that matter in 2026: Claude Pro with Claude Code, ChatGPT Plus with Codex, GitHub Copilot Pro, Cursor Pro and Grok 4.5. After that, we look at API prices, because as soon as you build AI into your own tooling, you no longer pay per month but per million tokens.

What are the three ways to let AI program?

Before you choose a subscription, it helps to know which of the three workflows suits you. They are not mutually exclusive, but you do pay for each one.

The chat. You paste code into a chat window, ask a question and paste the answer back. Slow, but unbeatable when you want to think out loud about an approach, do not understand an error message or want an unfamiliar piece of code explained. Every subscription in this article can do this.

The editor. An assistant that watches as you type, knows your project and suggests completions. This is GitHub Copilot and Cursor. It mainly saves time on routine work: boilerplate, tests, repetitive refactors.

The agent. You give an instruction at a higher level and the model works through multiple files on its own. Claude Code and Codex fall into this category. This is the most powerful and at the same time the riskiest: you have to review the result as if a junior developer had submitted it.

Claude Pro: the only subscription with a programming environment included

Claude Pro costs $20 per month, which after conversion and 21% VAT comes to €21.02, or €17.52 per month if you pay annually. That price includes Claude Code and Claude Cowork. That makes it the only subscription in the €21 class with a full programming environment included, and therefore effectively the cheapest of the bunch for developers.

The models behind it are Claude Sonnet 5 as the daily workhorse and Claude Opus 5 for heavy reasoning, both with a context window of one million tokens. That window is why Claude works well with large codebases: you can pass in a large part of a project at once without cutting and pasting.

Two honest caveats. The top model Claude Fable 5 is not included in Pro; in the apps, you pay for it with prepaid credits. And "unlimited" does not exist here either: you work with usage windows that you use up faster than you expect during long sessions on a large codebase. More about how Anthropic works is in Claude AI explained.

ChatGPT Plus with Codex: the broadest package

ChatGPT Plus also costs €21.02 per month and includes Codex, alongside GPT-5.6 Sol, Sora 2, GPT Image 2, Agent Mode and Deep Research. If programming is one of your tasks and not your whole workday, that is attractive: you pay once and get images, video and research on top.

GPT-5.6 Sol has a context window of 1.05 million tokens and received an update on 6 August 2026 with better factual accuracy and an effort slider that lets you set how much thinking the model puts into an answer. For code, that is practical: for a simple function you do not have to wait for an extensive reasoning process, for a tricky bug you do.

If you only want to chat about code and do not need an agent, you can get by with less. ChatGPT Go costs $8 (€8.41), and since 6 August 2026 the free version offers unlimited text chat on GPT-5.6 Luna.

GitHub Copilot and Cursor: help where you already work

The two editor tools are a different category from a chat subscription, and many developers use one alongside a chat.

GitHub Copilot Pro, at €10.51 per month, is the cheapest way to get AI into your daily workflow, and the jump from $10 to $39 for Pro+ is steep enough that you should first check whether you need it. Cursor Pro, at $20, is in the same price range as the chat subscriptions.

The calculation most developers end up making: Copilot Pro in the editor plus Claude Pro for the thinking comes to €31.53 per month together. That is manageable, but it is the first step toward the subscription stacking many people end up with.

Grok 4.5 and the open models

Grok 4.5 from xAI came out on 8 July 2026 and is strong at code, with access to live data from X on top. Its context window of 500K tokens is half that of Claude and OpenAI, which is noticeable with large codebases. The subscription is called SuperGrok and costs $30 (€31.53), making it more expensive than the competition without giving you a programming environment.

In addition, there is a whole layer of open models built specifically for engineering that you can host yourself or call via a cheap host:

• Kimi K2.7 Code from Moonshot (June 2026, 256K context) is made for long runs of coding agents.
• GLM-5.2 from Z.ai (16 June 2026, 1M context) is an open-weight engineering model.
• Qwen 3.8 Max from Alibaba (3 August 2026, 1M context) is the newest model from that stable.
• Muse Glimmer from Meta (August 2026), with 30B parameters under Apache 2.0, is small enough to run locally on a single consumer GPU.
• Llama 4 Scout and Maverick remain the standard if you self-host, with context windows of up to 10 million tokens.

If you do not want to send code to an American or Chinese cloud, running Muse Glimmer locally is the only option in this lineup where no data truly leaves your machine. An overview of all models is in all AI models at a glance.

What does AI for coding cost via the API?

As soon as you build AI into your own tooling, CI or scripts, the subscription disappears and you pay per token. These are the prices per million tokens, input and output.

Three things stand out. The output price is a multiple of the input price everywhere, and with code in particular you generate a lot of output, so that is where your bill is. Grok 4.5, at $2 and $6, has the most favorable ratio of the commercial top models. And the price difference between generations has become huge: GPT-5.6 Terra does roughly what last year's flagship GPT-5.5 did, but costs $1.25 and $10 instead of the $5 and $30 you pay for the current flagship. On 30 July 2026, OpenAI cut the price of Luna by 80% and that of Terra by 20%.

Which combination is right for you?

• You program daily and work from the terminal → Claude Pro (€21.02). Claude Code is included and one million tokens of context really makes a difference on large projects.
• You mainly want completions in your editor → GitHub Copilot Pro (€10.51). The cheapest serious option, and enough for routine work.
• Programming is one of your tasks, not your main task → ChatGPT Plus (€21.02). Codex plus images, video and research in one subscription.
• You are building something on top of a model → API, and start with Claude Sonnet 5 or Grok 4.5. Fall back on GPT-5.6 Terra or Luna for simple calls.
• Your code must not leave the building → run Muse Glimmer locally, or an EU option such as Mistral.
• You are a freelancer and book this as a business expense → enter your VAT number with the provider; how that works is covered in AI for freelancers and small businesses.

If you do not stick to one tool, the bill adds up quickly: Claude Pro, ChatGPT Plus and Copilot Pro side by side come to €52.55 per month. We bundle the models into one Dutch subscription of €12.95 per month, with invoices in euros. What we do not replace: Claude Code, Codex, Copilot and Cursor themselves. Those are tools, not models, and if you need them, you stay with the provider.

If you want to know how the models compare in other respects, see GPT-5.6 vs Claude 5 vs Gemini 3.5 or the complete comparison of AI subscriptions.

Conclusion