Reportedly: Huawei Ascend announces 0-day inference support for Moonshot AI's Kimi K3.1 Pro, a 2.8-trillion-parameter MoE model
Model: Kimi K3.1 Pro, Kimi K3 · availability, license and releases
Reported by IT Home and Pandaily · not yet confirmed by the company or a second independent outlet. We update this page when it is.
Huawei Ascend announced 0-day inference support for Moonshot AI's Kimi K3.1 Pro, a 2.8-trillion-parameter MoE model with native vision and a million-token context window. The support covers Ascend Atlas 800 A3, Atlas 900 A3 SuperPod, and Ascend 950 series. Moonshot AI released the full model weights on November 4, 2026, IT Home reported.
China context
- Original name
- 月之暗面
- Outside China
- Not stated in the sources yet
- Claims
- Company-reported; not yet independently evaluated
- For builders
- Developers outside China can study the model architecture and Ascend's optimization approach, but direct deployment on Ascend hardware may be limited by export controls and availability.
- For investors
- The 0-day support signals tight integration between Chinese model providers and domestic hardware, which may reduce reliance on foreign chips for large-model inference in China.
Translated from Chinese. Quotes and facts link to the original sources.
On November 4, 2026, Moonshot AI released the full weights of Kimi K3.1 Pro, the first model in the K3.1 series. Huawei Ascend announced same-day inference adaptation via vLLM Ascend and SGLang, enabling deployment on Ascend Atlas 800 A3, Atlas 900 A3 SuperPod, and Ascend 950 series. The model has 2.8 trillion parameters, native visual understanding, and a million-token context window. It uses hybrid linear attention with KDA linear attention layers and DSA sparse attention layers. Ascend optimized operators and frameworks for sparse attention, long context, and multimodal inference.
Kimi K3.1 Pro uses a hybrid linear attention architecture combining KDA linear attention and DSA sparse attention layers. Ascend's adaptation involved operator optimization for sparse attention computation, historical information filtering, preprocessing, and residual computation, plus distributed caching and framework capabilities for long-context inference.
Developers and enterprises using Huawei Ascend hardware can deploy Kimi K3.1 Pro on the day of its release, avoiding the usual delay for hardware-specific optimization. This reduces time-to-deployment for Chinese AI teams building on Ascend infrastructure.
For enterprises in China using Ascend hardware, immediate support for Kimi K3.1 Pro enables faster integration of a large multimodal model with long context for tasks like software development, enterprise knowledge processing, and financial research.
A verifiable next signal is whether Moonshot AI releases additional models in the K3.1 series and whether other hardware vendors (e.g., NVIDIA) provide similar day-0 support. Another signal is adoption of Kimi K3.1 Pro in enterprise deployments on Ascend hardware.