Weekly Recap · Week 25/2026

AI Recap: ChatGPT loses the majority, Anthropic's Korea drama — and the alternatives push to the front

A week that reshuffles the AI market: for the first time, ChatGPT no longer holds a majority, while Anthropic is caught between U.S. export policy and a rush of Korean conglomerates. At the same time, European and open alternatives are making themselves loudly heard — and AI delivers a genuine bright spot in medicine. The week of June 15–21, 2026 — in brief, with sources.

The week in one sentence

The AI market is fragmenting visibly: ChatGPT slips below 50% market share for the first time, Gemini and Claude are catching up, and after the Fable 5 ban, European (Mistral) and open, self-hostable models (MiniMax M3) move into focus. While Anthropic simultaneously fights and wins in Korea, OpenAI shows — with AI-assisted diagnosis of rare childhood diseases — what the technology can do at its best. In short: provider diversity has never mattered more than it does now.

ChatGPT slips below 50% market share for the first time

According to the "State of AI 2026" report from market researcher Sensor Tower, ChatGPT no longer holds a majority for the first time since it invented the category in 2022: 46.4% market share, while Google's Gemini climbed to 27.7% and Anthropic's Claude to 10.3%. By absolute users, ChatGPT stays out front with 1.1 billion — but Claude leads on paid conversion at ~13%. For businesses, that means the model choice is no longer "set it up once and forget it" — whoever can switch flexibly between providers has the edge. Which model is good for what, our AI tools comparison lays out.

Source: TechCrunch

Anthropic's Korea drama: export ban on one side, major customers on the other

The Fable 5 saga continues: SK Telecom was identified as the trigger of the U.S. export ban — the Korean conglomerate allegedly has China ties, according to the White House (SK Telecom denies this); on top of that, Amazon researchers reported vulnerabilities in Fable 5. In the middle of the crisis, Anthropic opened its Seoul office and held out the prospect of the models returning "in the coming days." Notably: despite it all, NAVER, Samsung SDS, LG CNS, Nexon and Hanwha announced major Claude rollouts on the very same day. Where the legal limits sit in all of this, our EU AI Act guide clarifies.

Source: Tom's Hardware Source: Korea JoongAng Daily

OpenAI's AI helps diagnose 18 rare childhood diseases

A research team from Boston Children's Hospital, Harvard and OpenAI used an OpenAI reasoning model (o3 Deep Research) to make 18 new diagnoses in previously unsolved cases — among them ten neurodevelopmental disorders. The AI provided leads that were then clinically validated by specialist physicians (published in NEJM AI). No replacement for doctors, but a tool that shrinks the pile of unsolvable cases — and a good example of how AI can do more than write text.

Source: OpenAI Source: NBC News

Mistral counters with "Vibe" and a push into industry

The French OpenAI challenger Mistral unveiled its product "Vibe" (the former "Le Chat") and announced a clear push toward industrial AI, complete with its own data-center buildout. For DACH companies, that's relevant: with a European provider, data stays inside the EU — an argument that carries more weight than ever after the U.S. export turbulence of recent weeks.

Source: DevX Source: Mistral

MiniMax M3 becomes the No. 1 open model

The Chinese lab MiniMax released the open weights and technical paper for M3 — making the model No. 1 among open-weight models (Artificial Analysis Index). It offers a 1 million token context, native multimodality and beats GPT-5.5 and Gemini 3.1 Pro in benchmarks — at just 5–10% of the cost. The real kicker for businesses: it's self-hostable. Run M3 on your own hardware and sensitive data stays in-house — a strong argument, especially after the Fable 5 ban. Note: commercial use requires a separate license from MiniMax.

Source: TechTimes Source: VentureBeat

Sources & further reading

As of June 21, 2026. Figures on market share, models and directives are per the companies, market researchers and the cited media, without warranty. Benchmarks and provider claims are, where flagged, reported by the vendors and should be verified independently.

Don't want to miss any AI development?

We keep an eye on the AI world for you — and translate what actually matters for your marketing, your tools and your automation.

Talk to us