GPT-5.6 vs Claude 5 vs Gemini 3.5: which model for which task?

In 2026, the question "which AI model is best" no longer has an answer. The three big families — GPT-5.6 from OpenAI, Claude 5 from Anthropic and Gemini 3.5 from Google — are close to each other on general tasks. The differences lie in what surrounds them and in the character of their answers.

That's why this comparison is organized not by model but by task. Six things people use AI for every day: programming, writing, research, translating into Dutch, working with documents, and making images and video. For each task you'll see who wins, why, and what it costs you.

How do the three families compare side by side?

First the hard data. Each provider has several models; these are the ones you'll come across in the subscription of around €21.

Two terms for anyone running into this for the first time. A token is a piece of a word; as a rule of thumb, a thousand tokens is about 750 words. The context window is how much text the model can keep in view at once: your question, your attachments, the earlier conversation and the answer combined. On that point the three families are practically equal: around a million tokens, or several thousand pages.

The API prices are what developers pay per million tokens. As a consumer you pay a subscription and never see those amounts, but they do say something about how heavy a model is. That Claude Fable 5, at $10 input and $50 output, is twice as expensive as GPT-5.6 Sol is no coincidence.

Programming: Claude, with Grok as a serious runner-up

In coding, it's long since stopped being about who types the nicest function into a chat window. It's about the tooling around it: can the AI read your files, run code, see the error message and correct itself.

Claude has gone furthest there. Claude Code simply comes with Claude Pro (€21.02 per month) and works independently in your project folder. Combined with Claude Opus 5, Anthropic's reasoning model, that's the strongest combination at that price for most developers. Reasoning here means the model first works out step by step what needs to happen before it starts writing, which makes the difference with complicated bugs.

Grok 4.5 from xAI (8 July 2026) is distinctly strong at code and, at $2 input and $6 output per million tokens, cheap to use. Its context window of 500,000 tokens is half that of the rest, though, which starts to pinch with large codebases. At €31.53 per month, its subscription is also the most expensive among the major providers.

GPT-5.6 Sol competes with Codex in ChatGPT Plus, and since the update of 6 August 2026 has an effort slider that lets you decide how long the model thinks. Handy: for a simple refactor you set it low, for a hard problem high.

If you self-host or work through the API, you have two more options: Kimi K2.7 Code from Moonshot is built for long runs of coding agents, and GLM-5.2 from Z.ai is an open engineering model with a context window of a million tokens. The full comparison of coding assistants, including GitHub Copilot Pro ($10) and Cursor ($20), is in coding with AI.

Writing: Claude, and it's not a close call

For writing in Dutch, the gap is the largest of all the tasks in this article. Our assessment of Dutch language quality:

The problem with GPT-5.6 isn't incorrect Dutch but Dutch that audibly reads as translated: expressions that come literally from English and an upbeat tone that's out of place in a Dutch business email. With Gemini it's the other way around: everything is correct, but there's no life in it, and whether you ask for a light or a formal text, you get roughly the same text back.

Claude does hold a requested tone to the end of a long piece. In practical terms: less rework. Why that is and what else Claude can do is covered in Claude AI explained; the full test with all nine providers is in which AI speaks the best Dutch.

Research: Gemini, with a caveat

Research covers two kinds of work: figuring out how something works, and checking whether something is true.

For the first, Gemini 3.1 Pro in Google AI Pro (€21.99 per month) is the strongest option, thanks to Deep Research: the model works through a series of sources on its own and delivers a report. Since 6 August 2026, GPT-5.6 Sol has had an update that makes it noticeably better at handling facts, and ChatGPT Plus has Deep Research too. In practice those two are close; Google wins on processing a lot of material at once.

The caveat is fundamental: none of the three gives you verifiable sources for every claim by default. If you do want that, you need a different kind of tool. That's the subject of Perplexity AI: the AI that actually cites its sources.

Translating into Dutch: Gemini overtakes Claude

This is the only task where Gemini is the best choice, and it's the exact opposite of the outcome for writing.

Translation is a task where you specifically don't want a voice of its own. You want the source text to come across accurately, with the same structure and without added interpretation. Gemini is good at that: correct, literal where it needs to be, and consistent in terminology across a long text.

Claude translates fine too, but tends to improve the text: smoother phrasing, more natural sentence structure. For marketing copy that's an advantage, for a legal or technical translation a risk. GPT-5.6 sits in between and in long texts sometimes slips back into English constructions.

Documents: the context window isn't the whole story

On paper the three barely differ: 1.05 million tokens at OpenAI and Gemini 3.5 Flash, 1 million at Claude and Gemini 3.1 Pro. In practice something else makes the difference, namely what you can do with those documents.

Google AI Pro has NotebookLM: you upload your own sources and Gemini answers only based on them, with a reference to the passage. For case files, study material and reports, that's a different way of working from pasting a document into a chat window. On top of that you get 2 TB of storage, which matters in practice with a large document collection. How that subscription is put together is covered in Gemini AI explained.

With its million tokens, Claude can handle large files just as well, but it runs into a different problem: the usage windows. Upload a few heavy documents in a row and you hit your limit sooner than you'd expect. ChatGPT Plus gives you 160 messages per three hours on the top model, which feels roomier for document work.

Images and video: only OpenAI and Google take part

Here Claude drops out completely: Anthropic makes no images and no video.

• OpenAI has GPT Image 2 and Sora 2, both in ChatGPT Plus. Video generation in a subscription of €21.02 per month is unique.
• Google has Nano Banana Pro for images and Veo 3.1 for video.
• Anthropic has nothing in this category.

For anyone who regularly creates visual material, this is often the deciding factor. And it's also the reason so many people keep ChatGPT Plus alongside another subscription.

What does it cost to pick the best per task?

All in all, the outcome of this comparison is awkward: there's no single winner, there are six winners.

So if you want to use the best model on all six tasks, you need three subscriptions: €64.03 per month, or €768.36 per year. That's the heart of the problem. Not that the models fall short, but that each sits behind its own paywall while almost nobody does just one kind of work.

If you do have to pick one subscription, look at what you do most often, not at what scores best on the hardest task. We weigh that up per type of user in ChatGPT Plus vs Claude Pro vs Google AI Pro, and every subscription type from every provider is laid out side by side in our comparison of AI subscriptions.

Conclusion