Ecosia Shifts AI Search Toward Open-Weight Models, Including Chinese LLMs

AI Author: EqualOcean News Updated 3 hours ago (GMT+8)

Ecosia, the Berlin-based search engine, is shifting more of its AI services toward open-weight models, including Alibaba’s Qwen, Z.ai’s GLM and Moonshot AI’s Kimi, through German AI platform Melious. Founder and CEO Christian Kroll told Politico that the change had roughly halved costs while improving quality and performance. The move shows how Chinese-developed models are entering Western products through European hosting and infrastructure.

ecosia

Kroll told Politico that Ecosia was disappointed with Mistral’s model quality and described it as about a year behind competitors. He also cited technical problems, including overloaded servers, and questioned whether Mistral was truly sovereign given its international investors. Ecosia had replaced OpenAI with Mistral in May. Kroll later clarified that the company’s focus is on open models, not their country of origin, and said it would probably use Melious, a European provider.

The transition is not a complete break with Mistral. Ecosia’s help page says the company currently uses several models side by side, including Mistral and open-weight models, routing different queries to different systems. It also says that when models are self-hosted, user questions and answers remain within Ecosia’s infrastructure rather than being sent to China or other regions outside the EU. The company has not published benchmarks or cost data behind Kroll’s claim of a roughly 50% reduction.

Founded in 2009, Ecosia uses its profits to fund tree-planting projects and has made privacy and sustainability central to its brand. Adding AI-generated answers to search creates new decisions about inference costs, energy use, data handling and the quality of results. Open-weight models give providers more flexibility to choose and run models on their own infrastructure, while still relying on technology developed elsewhere.

For Chinese AI developers, Ecosia’s shift illustrates one route to international adoption: their models can be embedded in foreign products without the developers operating those products directly. But the case is not simply a Chinese model replacing a European one. Ecosia’s approach combines models from different origins with European hosting, and it says it will select models based on performance, energy efficiency and data protection. The wider question is whether open-weight models can give European services greater control over deployment while leaving them dependent on capabilities developed abroad.


Contact Us:

Xingyiran@iyiou.com

yiran.png.png