Count the seven US assistants that matter. Claude works inside, behind or beside five of them. Only OpenAI’s ChatGPT and Meta’s Muse still run on nothing but their own engines. For Europe’s small firms, the name on the chatbot now tells you less than the name on the invoice.
FROM THE EDITORS
Every high street has a shop that sells five brands of beer brewed in the same plant. The labels differ, the bottles differ, the adverts argue with each other. Behind the wall, one brewery runs the tanks. Early on Wednesday, the American AI market got its own version of that brewery.
“Going forward,” Elon Musk wrote on X, SpaceX’s Grok bot “will use the best back end model for any given task, including Claude Opus 5.5, MidJourney, Suno and other leading APIs. Whatever is most likely to give you the best outcome.” The company that built Grok 4.5 and called it “Opus-class” now says it will call Opus itself when Opus is the better answer.
Five labels, one brewery
Line up the seven US assistants people actually name: ChatGPT, Gemini, Copilot, Meta AI, Grok, Claude and Perplexity. Then ask where Anthropic’s models sit in each one.
- Claude. Anthropic’s own front door.
- Copilot. Microsoft added Claude to 365 Copilot in September 2025 and in August 2026 began rolling out Claude Opus 5 through the model selector in Word, Excel, PowerPoint, Copilot Chat, Copilot Cowork and Copilot Studio.
- Perplexity. Claude Opus 5 is a selectable model for Pro and Max subscribers, and an orchestrator inside Perplexity Computer.
- Grok. As of this week, one of the engines SpaceX’s bot will route to.
- Gemini’s house. The honest asterisk. Claude does not power Gemini. But since 6 October it works inside Google Docs, Sheets and Slides as an add-on, in the same sidebar where Gemini lives. Not behind the bar, but at the bar.
That leaves two holdouts. OpenAI runs ChatGPT on its own models. Meta rebuilt Meta AI around Muse Spark, the first model from its Superintelligence Labs, released in April. Both are betting that owning the whole bottle, label and brew alike, is the business. Everyone else has decided the label is the business and the brew can be bought.
Why a rival would rent the rival
The obvious objection: surely Musk is just hedging, and Microsoft just keeping OpenAI honest. Both are true, and neither weakens the point. A hedge announced in a post with five million views is no longer a hedge. It is a product decision. Users of an assistant do not care whose weights answered. They care whether the answer was right. Once one large assistant routes openly to the best engine, the others have to explain why they don’t.
That turns the market inside out. For three years the race was framed as model against model, a league table of benchmarks. Increasingly it is storefront against storefront, with a small number of breweries supplying the shelves. On Pew’s February count, Claude was used by 6 per cent of American adults, behind ChatGPT at 44 and Gemini at 24. Measured by front door, Anthropic is a small shop. Measured by engine room, it is in most of the street.
What it means for a firm in Ghent or Gdańsk
For a European small business, this is mostly good news and partly a paperwork problem.
The good news: you no longer have to pick a model to get a good one. The assistant your staff already open, whether in Excel, a search tab or a social app, will increasingly reach for whichever engine fits the job. The lock-in moves from the model to the workflow, and workflows are yours.
The paperwork: a routed request passes through two companies. The one whose logo you see, and the one whose model did the thinking. Under GDPR that second company is a processor too, and it belongs in your records. “We use Copilot” used to be a complete answer to a customer asking where their data goes. It is now half an answer.
What to do this week
- Read the label behind the label. For each assistant your team uses, find out which models it can route to, and whether you or an admin can switch that off.
- Update the processor list. If Copilot can call Claude, both Microsoft and Anthropic belong in your data records.
- Keep customer data on business plans. Consumer tiers of any of these assistants are not the place for names, invoices or HR files.
- Judge the answer, not the badge. Run the same real task, a quote or a reply to a complaint, through two assistants. Pick the one your staff will actually open.
For a while it looked as though the AI wars would end with one winner. They are ending with something more familiar: a crowded shelf, loud labels, and a short list of breweries behind the wall. Two of the big names still brew their own. The other five now stock the same brewer somewhere on the shelf.
Sources
- Elon Musk on X, 7 Oct 2026 — https://x.com/elonmusk/status/2107724314451878104
- ProPakistani, “Microsoft Now Supports Claude Across Its Copilot Apps”, 28 Aug 2026 — https://propakistani.pk/?p=1072049
- Digg / Perplexity, “Perplexity Adds Claude Opus 5 to Platform and Computer” — https://digg.com/tech/hhc52pv3
- Signal Digital, “The Rival Just Got a Desk in Google’s Office”, 6 Oct 2026 — https://signaldigital.net/2026/10/06/the-rival-just-got-a-desk-in-googles-office/
- Axios, “Meta debuts Muse Spark, first AI model under Alexandr Wang”, 8 Apr 2026 — https://www.axios.com/2026/04/08/meta-muse-alexandr-wang
- Axios, “Musk’s SpaceXAI releases new model, Grok 4.5”, 8 Jul 2026 — https://axios.com/2026/07/08/spacexai-grok-new-model
- Pew Research Center, “ChatGPT tops the list of chatbots Americans use” (survey 17–23 Feb 2026) — https://www.pewresearch.org/chart/chatgpt-tops-the-list-of-chatbots-americans-use-across-age-groups/
Leave a Reply