The latest in AI: what you need to know to catch up

That pace makes choosing hard. Every week brings a new release, a new benchmark or a new promise. In our webinar Catching up on AI, we spend one hour laying out what changed, which tools make a real difference today and how to make a choice you can still stand behind a few months from now. In this blog, we give you a sneak peek.

A new generation of models, at record speed

Anthropic, OpenAI, Google and xAI all released new models this summer. Claude got Sonnet 5, Fable 5.1 and Opus 5.5. OpenAI launched GPT-6 Astra, followed by GPT-6 Sol and Luna. Google shipped three versions of Gemini Flash in six weeks.

On 22 September, Claude Opus 5.5 and GPT-6 Sol and Luna even came out within minutes of each other, both cheaper than their predecessors. That says a lot about the market: the big players are keeping a close eye on each other, and prices are dropping while quality goes up.

For anyone looking to use AI in their business, the lesson is simple. If you wait for the best model, you’ll wait forever. There’s always a next version. So the real question is which model fits the work you want to do today.

From best model to best fit

At the top of the rankings, the leading models are now very close to each other. Prices, however, vary widely. The most expensive model costs up to thirteen times as much as a fast model that handles many tasks just fine. Each model family also has its own character, and that character stays remarkably consistent from one version to the next:

• Claude (Anthropic) is strong at writing, long documents and work where the model operates tools itself. It follows instructions closely.
• ChatGPT (OpenAI) has the broadest ecosystem and is strong at reasoning, images and voice. Its range of variants is large, though, and not always easy to navigate.
• Gemini (Google) has a very large context window and works well with PDF, audio and video. The Flash models are fast and cheap.
• Mistral is a European company that offers hosting in the EU at a competitive price. On complex reasoning tasks, it still trails the leaders.
• Open models such as DeepSeek, Qwen and Llama can be self-hosted and kept fully under your own control. Management, security and compliance are then your responsibility.

Once you understand where each family shines, you don’t need to follow every release.

Agents move from testing to real work

The biggest shift this summer may not be in the models themselves, but in how we work with them. With Claude Cowork and ChatGPT Work, you hand over a task and the AI works through it step by step in your own files and systems. You get a result back instead of an answer in a chat window.

Scheduled tasks now also run on a server. You can close your laptop and the work carries on. That opens the door to longer jobs: working through a case file, preparing a report, pulling information from dozens of documents.

It does ask something of whoever gives the instructions. An agent working on its own can’t stop to check something with you. The sharper the brief, the more useful the result. And work that runs by itself is easily forgotten. So agree on who reviews the output.

Today, agents work reliably for:

• multi-step tasks with a clear end result, such as research or a report;
• work within one clearly defined environment, such as your documents or your mailbox;
• connecting to your own tools through connectors;
• work that someone can quickly review afterwards.

Where things still go wrong: tools running unsupervised for long periods, tasks where mistakes have financial consequences or affect client relationships, automating a process nobody has ever written down, or granting broad access rights because it’s easier. So start with work that takes time but carries little risk.

AI is already in the tools you use

A second trend is getting less attention. AI is moving into the software you already use. In Notion, a good AI instruction becomes a button the whole team can reuse. In Zapier, you describe in one sentence what needs to happen, and an AI bot builds the integration. Gemini is built into Google Sheets and Gmail. HubSpot has Breeze agents for prospecting, content and service, working with your own CRM data.

That has a practical consequence. Many companies are already paying for AI they don’t use. So check what’s included in your existing software before you sign up for a new subscription.

The AI Act: transparency is now mandatory

Since 2 August 2026, the transparency obligations of the European AI Act apply. If you use a chatbot, you must make it clear that people are talking to AI. Generated or manipulated images, audio and video must be labelled. Part of the regulation already applied earlier: since February 2025, practices such as social scoring and emotion recognition in the workplace are banned, and employees must have a sufficient understanding of what the tools they use actually do.

Most SMEs use AI but don’t build or sell it. That spares them a lot of obligations. It doesn’t mean there’s nothing to do. In the webinar, we go through six checks for your own business:

1. Take stock. Which AI tools do you use, for what, and with which data? The list is usually longer than expected.
2. Define your role. Are you a user or a provider of AI? That determines which rules apply to you.
3. Be transparent. Let customers know when they’re talking to a chatbot or looking at generated images.
4. Build AI literacy. An internal session and clear agreements are often enough.
5. Know what’s banned. Some applications simply aren’t allowed.
6. Check your contracts. Where does your supplier store your data, and is it used to train models?

Please note: this is not legal advice. Have a legal professional review your specific situation.

What else you’ll take away from the webinar

New models, agents and rules shape the landscape after the summer. But in our webinar, we go one step further, to the question that matters most for your team: which tools belong in your work, and how do you choose them?

Here’s what to expect:

• The landscape, mapped out. We split the market into seven categories, from your everyday workhorse to automation and building your own. That way you see where your team stands today and where the biggest gains are.
• Three questions to help you choose. A simple way to pick a tool without testing for a year.
• Which developments you can safely ignore. At least as important as what you do act on.
• Same question, three tools. We gave Claude, ChatGPT and Perplexity exactly the same task. The differences are bigger than you’d think, and they say a lot about how best to use each tool.
• A recipe to get started right away. Four steps to try a new tool this week on a task from your own schedule.

You’ll leave with a shortlist for your own work and a concrete first step. The webinar is for anyone who wants to get started with AI or is already using it, in marketing, HR, finance or beyond. No prior knowledge needed.

When: 22 October 2026, 10:00

Where: online

Sign up: https://www.unikoo.be/events/webinar/bijgepraat-over-ai-het-landschap-na-de-zomer-van-2026

‍