Search Engine Ecosia Swaps Mistral for Chinese A.I.
Trending Topics Jakob Steinschaden
Ecosia dumped Mistral and switched to Chinese open A.I. models. Europe’s “sovereign A.I.” story just got awkward: the best open models are still mostly Chinese.
Based on reporting by Trending Topics, Jakob Steinschaden — read the original for the full story.
Summary, retelling and take written by AI under human oversight; images are AI-generated illustrations. How we work · Report an error
Europe’s A.I. sovereignty push has hit a very public snag in Berlin. Ecosia, the search engine used by public authorities as well as regular users, has dropped Mistral as its A.I. supplier and moved to open models instead, including Alibaba’s Qwen, Z.ai’s GLM and Moonshot AI’s Kimi, Politico reports.
The switch lands at a bad moment for Mistral. The French company this week unveiled Large 4, its most powerful model yet, but Ecosia’s chief executive, Christian Kroll, says the company still found Mistral’s models too far behind the competition. He told Politico they lag by about a year, and reports also point to frequent server overloads. Ecosia had only switched from OpenAI to Mistral in May, framing that move as an act of European independence.
Now it has gone the other way. According to Ecosia’s help page, its search summaries had most recently run on Mistral Small 3.2, while its A.I. chat used Mistral Small 4. The new setup routes its models through Melious, a German platform that runs open A.I. models on servers in eight E.U. countries. Melious says the data never leaves Europe, and its lineup includes Z.ai’s GLM 5.3, Moonshot AI’s Kimi K3 and DeepSeek V4.1 Flash.
The economics seem to have helped the decision. Reports on the change say Ecosia cut the cost of its A.I. services by about half while improving performance. Kroll also questions whether Mistral really counts as sovereign, given that the French company relies on money from international investors. That is the ugly little twist here: a European search engine trying to buy European independence ends up finding better open models from China.
Mistral is not exactly surrendering. Large 4 was trained in European data centers and will get open weights at the end of October. It scores 38.4 on the Intelligence Index from Artificial Analysis, far above Mistral’s older models, though still behind every Chinese model ahead of it in the ranking. Price is not a clean win either: Large 4 sits at $1.13 per task in the index, cheaper than some rivals but nowhere near the $0.27 shown for DeepSeek V4.1 Flash.
My take — AI-written commentary, not fact-checked reporting
This is the part of European A.I. policy nobody likes to say out loud: sovereignty slogans are cheap, good models are not. If the best open systems come from Chinese labs and the European alternative trails them, the market will do what markets do and ignore the flag on the box.
Read more about this at: Trending Topics
