Updates·October 19, 2026, 21:36

AI tools are getting more numerous, cheaper and moving into the workday

AI-generated from the sources listed, with automated source checks.

This week's updates show two trends: big companies are building AI into everyday software, while open and cheap models, especially from China, are pushing hard. Many of the figures come from the companies themselves, though.

AI-generated image

Looking at this week's updates as a whole, two tracks emerge. One is that AI is becoming a permanent part of the programs we already use at work. The other is that more and cheaper models are arriving that you can download and run yourself.

AI moves into the workday

In ChatGPT, the Chrome extension can read across open tabs in Chat mode, and in Work mode it can also fill in forms, send messages and download files. According to Business Today, the controls require the ChatGPT desktop app, and it is unclear whether they apply to all plans. OpenAI has also introduced Slides, so teams can build presentations together with AI, but pricing and Danish terms have not been disclosed. And the voice feature has gained 13 new voices, including in Hindi, French and German, which are meant to sound more natural.

Microsoft is taking a different route into the office: more than 20 legal features in Copilot, including contract review in Word. They are aimed at customers with commercial Copilot licenses, and pricing has not been disclosed. Adobe, meanwhile, lets Firefly users choose from more than 10 AI models from other companies, including Google, OpenAI and Runway, and promises that users' material will not be used to train the models.

It is also worth noting that older models are disappearing. OpenAI is removing GPT-4o, GPT-4.1 and the mini versions from the API (developers' access to the models) on January 14, 2027, and recommends gpt-6-luna as a replacement. If your company or a vendor uses the old models in custom-built solutions, there is a deadline to keep an eye on.

Open and cheaper models push forward

China's DeepSeek has released a new version of its Flash model and, for the first time, opened its own search engine to developers. The figures for the model's capabilities come from DeepSeek itself and have not been independently verified. The company promises better coding at a lower price, but independent tests are lacking. According to Chinese media, OpenCode has made DeepSeek V4.1 Flash free, but the free version can only remember 200,000 tokens (small chunks of text) at a time.

Others are opening their doors too. Meta has released the weights for Muse Spark Open, so developers can use it freely under certain terms. In Meta's own comparison, it trails several Chinese models in coding but leads in tool use. Black Forest Labs says a version of the open image model FLUX 3 Image Klein can run on a graphics card with 8 GB of VRAM.

What can you do?

The pattern is that almost all the impressive figures come from the companies themselves. That also goes for this week's new comparison site, where the company behind it sells access to the models itself. So treat the news as promises, not as the final word.

The two tracks also point in different directions. The big services make things easy, but tie you to their terms. The open models offer freedom, but require technical work, and according to Developers Digest, there is still a gap to the best models on long-running tasks for China's Step 5 Preview.

In practice, you can check whether your tools use models that are being shut down, and try new features on non-critical material before they touch contracts or customer data. Also watch whether independent tests confirm DeepSeek's promises.

Source

More on this topic

Get the week's AI news in your inbox

Choose your level, topics and length. One email a week, unsubscribe at any time.

Subscribe to Promptly Newsletter