Huawei Ascend adds day-zero inference support for Kimi K3.1 Pro
Huawei announced that Ascend completed same-day inference adaptation for Kimi K3.1 Pro through vLLM Ascend and SGLang, according to IT之家 citing the Huawei Computing account. The vendor says deployment is supported on Ascend Atlas 800 A3, Atlas 900 A3 SuperPod, and Ascend 950 supernodes, with optimizations for sparse attention, million-token context, and multimodal inference. The 2.8-trillion-parameter model was released with full weights on November 4, 2026.
The day-zero adaptation gives developers an immediately available Ascend deployment path for the newly released Kimi K3.1 Pro, while the reported framework and operator optimizations address serving a large MoE model with long context.
#Kimi K3.1 Pro#Huawei Ascend#Moonshot AI#vLLM Ascend#SGLang#Atlas 800 A3#Atlas 900 A3 SuperPod#Ascend 950