Here’s the deal, folks. On May 13th, at the Create 2026 Baidu AI Developer Conference, Shen Dou, Executive Vice President of Baidu Group and President of Baidu Intelligent Cloud, spilled the beans. The Kunlun Chip P800 has completed large-scale verification and has been shipped to multiple clusters since 2025. Honestly, that’s a big deal.
But wait, there’s more. On the Kunlun Chip nationwide cluster, they’ve successfully trained the Wenxin 5.1 major version, with an effective training rate of 97%. The cluster’s linear scalability exceeds 85%, meeting the requirements for large-scale training of cutting-edge models, including computing precision, operator stability, framework adaptability, and long-cycle running.
And, let’s talk about the Tianchi 256-card super node, based on the Kunlun Chip. It was lit up last month and will be officially launched in June. The throughput performance has increased by 25% compared to the previous generation, and it’s completed the adaptation of mainstream models like Wenxin, DeepSeek, GLM, and MiniMax. The inference efficiency has been boosted by 50%, and the network architecture has been upgraded to HPN5.0, with a 50% reduction in end-to-end latency. Plus, it supports the on-demand construction of massive clusters with tens of thousands to millions of cards.
