Case Study: MiniMax achieves real-time AI search and trillion-scale deduplication with Zilliz Cloud

MiniMax hits 30ms latency at 5,000+ QPS with Zilliz

Minimax, a leading provider of large language models, faced significant infrastructure challenges as its AI platform, Talkie, scaled to tens of millions of users. Their Redis-based solution could not handle the low-latency demands of real-time recommendations, and traditional systems made trillion-scale training data deduplication economically impossible. They turned to vendor Zilliz and its product Zilliz Cloud for a purpose-built solution.

Zilliz implemented a unified platform using Zilliz Cloud to address both challenges. For real-time recommendations, its native vector operations delivered under 30ms latency at over 5,000 QPS. For data deduplication, a custom MinHash + LSH engine built into Zilliz Cloud made processing 2x faster and reduced costs by 3-5x. This resulted in a significant drop in infrastructure spend, improved model quality, and simpler operations for Minimax.

Minimax