Ecosia Switched to Chinese AI Models for Search
The search engine shifted to top-tier open models hosted on E.U. servers to cut costs and improve reliability.
Updated on Oct. 8, 2026 in Artificial Intelligence
Live Poll
Should European companies prioritize using local A.I. models over higher-performing models from other countries?
Ecosia has replaced its previous AI provider, Mistral, with a suite of Chinese models including Alibaba’s Qwen, Z.ai’s GLM, and Moonshot AI’s Kimi. The search engine is accessing these models through the German hosting platform Melious.
Why it matters
The migration reflects an ongoing shift toward integrating top-tier open-weight models to satisfy search-engine demands for both quality and consistent server reliability. Ecosia’s decision prioritizes lower operational costs while addressing previous performance concerns.
The switch to Melious halved overall AI service costs, leveraging models like DeepSeek V4.1 Flash at $0.27 per task. This compares to the $1.13 cost per task of Mistral’s Large 4, which currently holds a 38.4 score on the Artificial Analysis Intelligence Index.
The players
Ecosia
A Berlin-based search engine company that uses its revenue to fund tree planting and is shifting its AI infrastructure.
Mistral
A French AI laboratory building large language models and transitioning into a cloud-infrastructure provider.
Melious
A German platform that provides server infrastructure for AI model hosting across the European Union.
Alibaba
A major Chinese technology conglomerate that develops the Qwen series of large language models.
Z.ai
A Chinese AI developer that creates the GLM model series and is currently designated on the U.S. entity list.
The details
Ecosia now routes search queries through Melious, a German platform that distributes model hosting across servers in eight E.U. countries. By shifting away from Mistral, the company aims to move past persistent reliability issues. Note that Z.ai, one of the new model providers, has been on the U.S. entity list since 2025.
Timeline
2025: Z.ai was placed on the U.S. entity list.
May 2026: Ecosia transitioned from OpenAI to Mistral models.
Summer 2026: Mistral began its transition into a neocloud entity.
October 2026: Mistral is scheduled to release open weights for its Large 4 model.
The Tech Race
Ecosia’s decision to integrate models from Chinese developers follows the tightening of international trade restrictions and the U.S. entity list. This move highlights a growing divide as search engines evaluate the performance of non-Western open-weight models against established providers.
Users of the Ecosia search engine will interact with AI summaries and chat features powered by the new Chinese model suite immediately. This infrastructure change is intended to maintain service availability while reducing the company's reliance on higher-cost providers.
The takeaway
Ecosia has prioritized operational cost and local server reliability by diversifying its model stack away from Mistral. Observers should track Mistral’s performance after the scheduled October 2026 release of its open-weight Large 4 model to see if it regains lost enterprise footing.
What happens next
Mistral is expected to release open weights for its Large 4 model at the end of October 2026.
Further reading
For more on the changing landscape of model integration, read our Artificial Intelligence coverage.
Source note: This article includes information reported by Trending Topics.
Live Poll
Should European companies prioritize using local A.I. models over higher-performing models from other countries?
