Alibaba reduced the cost of training AI search engines by 88%

Chinese tech giant Alibaba has announced a breakthrough in reducing the cost of training AI models focused on search. This is reported by SCMP.

Researchers have introduced a new approach called ZeroSearch. It allegedly improves the search capabilities of neural networks by 88% through simulations and without interacting with real systems like Google.

The method allows to avoid high costs often associated with routing requests through commercial tools. Models already trained on extensive knowledge bases generate "quality content" in response to queries.

It is claimed that ZeroSearch transforms the reference model into a search engine capable of training other AI systems to respond to queries. The technology reduces reliance on expensive external search infrastructure.

For example, sending 64,000 requests to Google via the API cost $586.7. Generating responses for training using an AI model with 14 billion parameters cost $70.8.

Innovation can expand search capabilities, researchers noted.

Recall that in April, the Chinese tech giant released a new family of AI models Qwen3, which are "capable of matching or even surpassing in some cases" the best solutions from Google and OpenAI.