Ecosia drops Mistral: "A year behind", now the search engine bets on open models

The search engine Ecosia, which uses its profits to plant trees, is dropping the French AI company Mistral AI. Instead, it will use open AI models, which also include Chinese models. That's according to Heise, Cybernews and EU-Startups, among others.

The switch comes quickly. In May 2026, Ecosia replaced the American OpenAI with Mistral as the provider for its AI features. Now, in October, Mistral itself is being replaced.

Why is Ecosia switching?

According to Ecosia's founder and CEO, Christian Kroll, Mistral is a year behind its competitors. The media also mention technical problems and overloaded servers. According to Heise, Ecosia is such a large customer that Mistral's infrastructure could not keep up, which led to regular outages.

Kroll also says that the switch has halved costs, while quality and performance have improved. He also mentions that there is a lot of nuclear power in France's energy mix, which does not fit well with Ecosia's climate mission.

He also questions how independent Mistral really is from the US. The company has received money from investors such as Andreessen Horowitz, Nvidia and Salesforce Ventures.

What are open models?

An open model (often called open-weight) is an AI model whose internals the developers have released. Others can therefore download it, adapt it and run it on their own servers instead of renting access from a single company.

Ecosia mentions models such as Qwen (from Alibaba), GLM (from Z.ai) and Kimi (from Moonshot AI). They are run via the German platform Melious.ai on European infrastructure with renewable energy. Data is thus processed in Europe.

Criticism and caveats

Kroll stresses to EU-Startups that the choice is about open models and not about China. He points to cheaper operation, flexibility and lower energy consumption. On politically sensitive topics such as Taiwan, according to him, Ecosia can use other models and is working to remove political bias in the answers.

Experts are not entirely convinced. According to Heise, Rasmus Rothe from the German AI association warns that political biases in training data run deep and are hard to remove afterward.

What does it mean for you?

If you use Ecosia, you may over time see different answers in the AI chat and in the AI overviews. Ecosia promises better quality. Whether the Chinese models bring a skewed angle on political topics is still an open question.