Last updated: 10 July 2026
This is an article with a short shelf life, and that is deliberate. The AI field is moving so fast now that "the best models" is a perishable product. What holds today is partly out of date in three months. So read this as it is meant: a snapshot of where the landscape stands in July 2026, not a timeless answer key. The point is not to crown a winner, but to give an honest picture of what exists, and what it actually means for a Norwegian business.
The big three, and a remarkable couple of weeks
The summer of 2026 has been one of the most dramatic periods in the field's history, for reasons that are as much about politics as about technology.
Anthropic (Claude). Anthropic's most powerful widely available model, Claude Fable 5, launched on 9 June. Just three days later, on 12 June, US authorities imposed export controls that forced Anthropic to temporarily shut off access for all users. The controls were lifted on 30 June, and the model returned on 1 July. It was the first time a frontier AI model was switched off by government order. At the same time, on 30 June, Anthropic launched Claude Sonnet 5 as the new default model for ordinary users.
OpenAI (GPT). OpenAI followed a strikingly similar pattern. GPT-5.6, in three variants called Sol, Terra and Luna, was first made available on 26 June to a small group of approved partners, gated behind a government security review. Yesterday, 9 July, the models became widely available. The trigger for the review was the same in both cases: the models' increased capabilities in cybersecurity.
Google (Gemini). Google's flagship is Gemini 3.1 Pro, with a successor (3.5 Pro) announced as "coming". Google's strength lies less in leading the benchmark race and more in the integration. Gemini is tightly woven into Gmail, Docs, Drive and the rest of the Google ecosystem, which makes it a natural choice for businesses already working there.
The common denominator in the summer's events is worth noting: governments have started treating the most powerful AI models as something close to strategic technology. Both of the major American providers were affected within the same month.
What is happening quietly: the open models
While the headlines are about Claude and GPT, another race is going on that deserves attention: the open models, meaning models where the "weights" are freely available and which can be run on your own hardware.
Here Chinese players in particular, such as DeepSeek, Qwen (Alibaba), Kimi (Moonshot) and GLM (Z.ai), have taken big steps. In several areas (code generation, resource efficiency, price) they now compete with, and on some tasks surpass, Western closed models. On the Western side Meta's Llama is still the best-known open model, while Google's Gemma offers a free variant with a generous licence.
What makes the open models interesting for businesses is not primarily the benchmark numbers. It is two other things. One is price (they are often dramatically cheaper per use than the closed frontier models). The other is control (they can run on your own servers, so that sensitive data never leaves the business). For certain uses, particularly where data protection weighs heavily, that is decisive.
What this actually means for a Norwegian business
It is easy to drown in model names and benchmark percentages. For the vast majority of businesses that is irrelevant. Here is what actually matters:
You rarely need the most powerful model. The frontier models get the headlines, but they are also the most expensive. For most practical tasks such as customer service, text drafts, summarising and simple automation, a cheaper model does the job excellently. The most expensive model is almost never the right choice for daily use.
The choice is more about tooling than about the model. How the AI is connected into your workflow (which data it has access to, how it is triggered, what it is allowed to do) usually matters more for the result than which underlying model is used. We have written about what AI can actually do at work, and that is where the real value lies.
Data is still the crux. Whichever model you choose, the question of which data it gets to see, and where that ends up, is the most important one. That is extra relevant now that some models can be run locally and others send everything to an external provider. We have a separate article on AI and your data that goes into this more thoroughly.
Do not lock yourself in. When the landscape changes this fast, it is unwise to build everything around one particular model from one particular provider. The summer's events, where frontier models were switched off and on by government order, are a reminder that access is not guaranteed. A solution that can switch model without being rebuilt is a safer solution.
So where do we stand?
Briefly summarised in July 2026: the Western closed models (Claude, GPT, Gemini) still lead on the heaviest tasks, but the gap is narrowing. The open models are closer to the frontier than ever, dramatically cheaper, and a real alternative where price and data control count. And for the first time, governments have started intervening in who gets to use the most powerful tools.
For a business the practical lesson is the same however the race ends: choose tools according to need, not according to headlines, keep control of your data, and build so that you can switch when the landscape changes again. Because it will, probably before this article has had time to grow old.
Does your business need help working out what AI can actually do for you, without the hype? Feel free to have a no-obligation talk.