Xiaohongshu released the open-source dots3-note preview; Ascend completed support on Atlas 800 A3 and Atlas 900 A3 SuperPoD.

Xiaohongshu released the open-source dots3-note preview model on August 14, 2026. The supplied source did not provide the model’s parameter count, architecture, training corpus, license terms, benchmark results, or intended workload.

Huawei’s Ascend ecosystem completed full adaptation of the model on the Atlas 800 A3 and Atlas 900 A3 SuperPoD systems. The Atlas 900 A3 SuperPoD is identified as a supernode system rather than a single accelerator.

Deployment and inference are provided through the open-source vLLM Ascend inference engine. The source said performance optimization was continuing but supplied no latency, throughput, token-rate, precision, or cluster-scale figures.