Published: October 8, 2026 · Last updated: October 8, 2026
TL;DR: Ecosia, the Berlin-based search engine that funds tree planting with its ad revenue, has stopped running its AI search on Mistral and switched to open-source models, including ones built in China, Politico reported Thursday, citing Mistral’s disappointing output quality. The timing stings: it comes a day after Mistral said its newest flagship model beats the best Chinese open models on multiple benchmarks.
Politico reported Thursday that Ecosia, the German search engine known for planting trees with its ad profits, has quietly dropped Mistral as the AI engine behind its search summaries and chat features. In its place: a mix of open-source models, including ones developed in China. Politico’s sourcing points to a blunt reason for the move: Mistral’s output wasn’t good enough for what Ecosia needed, even though both companies are European and have leaned on language about data independence from Silicon Valley.
Ecosia isn’t a bit player. The company has run a Bing-powered search engine out of Berlin since 2009, building a loyal user base that likes seeing a cut of every ad click go toward reforestation projects. Like most search engines chasing the AI wave, it layered generative features on top of its results over the past couple of years, including AI-written summaries and a chat mode meant to save people a click. Politico’s report notes that Ecosia counts some German government services among its users, which is part of what makes a switch away from a French-made model notable. Mistral has spent two years positioning itself, with backing from the French government and EU officials eager for a homegrown alternative to OpenAI and Google, as exactly the kind of AI a European public-facing product should trust.
That positioning took a hit the day before Politico’s story ran. On October 7, Mistral released Mistral Large 4, nicknamed “Le Chonk,” a 1.05 trillion parameter open-weight model the company said outperforms leading Chinese open models on several benchmarks. We covered that launch here. Benchmark wins are Mistral’s strongest argument against both the Chinese labs eating into its market and the broader perception that European AI lags behind. Ecosia walking away from Mistral’s models for real-world quality reasons, in the same week Mistral claimed a benchmark win over China, undercuts that argument in a way no leaderboard screenshot can fix.
Politico’s report doesn’t name every model Ecosia moved to, and Ecosia hasn’t published a detailed breakdown of its new stack. But the direction is clear enough: open-weight models anyone can download, self-host, and fine-tune, rather than a paid relationship with a single lab’s API. That category increasingly includes Chinese models like DeepSeek’s and Alibaba’s Qwen family, which have drawn a growing number of Western companies simply because they perform well relative to their cost and licensing terms. DeepSeek’s own lab is reportedly chasing a $12 billion funding round right now, which says something about how much demand its models have found outside China.
There’s a tension worth sitting with here. A big part of the pitch for “sovereign AI” in Europe, the reason Mistral gets French government money and a seat at EU policy tables, is that European governments and companies shouldn’t have their search queries and institutional data running through foreign-controlled models they don’t govern. Ecosia choosing Chinese open-weight models instead of Mistral’s doesn’t solve that problem. It just swaps which foreign dependency people worry about. If European AI products can’t deliver quality that beats both the US labs and the Chinese ones on home turf, the sovereignty argument starts to sound like a subsidy pitch rather than a product advantage.
Worth flagging: Politico’s full reporting sits behind a subscription wall, and the exact scope of Ecosia’s switch, including which products and how permanent the change is, isn’t fully public yet. Treat the broad strokes here as reported by Politico and relayed by other outlets, not as confirmed firsthand by either company.
Related: Mistral’s new Large 4 model claims it beats China’s best open models, and DeepSeek is reportedly chasing a $12 billion round toward a 2027 IPO, two sides of the same race this story sits in the middle of.
Bottom Line: A European search engine swapping a European AI lab for Chinese open-source models, in the same week that lab claimed a win over China, is a bad week for the “buy European AI” argument. Benchmarks are cheap to win. Production quality, the thing that actually made Ecosia switch, is not, and that gap will keep showing up in decisions like this one long after the leaderboard screenshots stop mattering.
Follow Teck Hustlers: Facebook · X · Instagram · LinkedIn · TikTok · Pinterest
