AnswersPublished18 sources
Alibaba Cloud’s M890 SuperPoD Adds Day-Zero Support for Kimi K3.1 Pro
Alibaba Cloud says its M890 SuperPoD completed day zero adaptation of Kimi K3.1 Pro weights and end to end validation with vLLM and SGLang; Zhongwei invitation testing also opened.
Published byEdited with GPT-6 LunaImages generated with GPT Image 2
Research answer
Key takeaways
- Alibaba Cloud says its M890 SuperPoD completed day zero adaptation of Kimi K3.1 Pro weights and end to end validation with vLLM and SGLang; Zhongwei invitation testing also opened.
AI Prompt
openai.comCreate a landscape editorial hero image for this Studio Global article: What did Alibaba Cloud’s November 5, 2026 announcement, reported by IT Home, reveal about Lingjun Zhenwu M890 SuperPoD’s day-zero adaptation. Article summary: Alibaba Cloud’s November 5 announcement said Lingjun Zhenwu M890 SuperPoD had completed day-zero adaptation of Kimi K3.1 Pro’s weights and end-to-end validation with both vLLM and SGLang, while opening invitation testing. Topic tags: general, general web, documentation. Style: premium digital editorial illustration, source-backed research mood, clean composition, high detail, modern web publication hero. Use reference image context only for broad subject, composition, and topical grounding; do not copy the exact image. Avoid: logos, brand marks, copyrighted characters, real person likenesses, fake screenshots, UI text, readable text, watermarks, charts with fak
Alibaba Cloud says its Lingjun Zhenwu M890 SuperPoD adapted Moonshot AI’s Kimi K3.1 Pro weights on day zero and completed end-to-end validation with vLLM and SGLang. The November 5 announcement also opened invitation testing for SuperPoD instances in Zhongwei. Published framework recipes and an Alibaba Cloud deployment guide provide practical evidence of support, though they do not establish independently measured performance or production reliability. 3
17
26
27
What the day-zero announcement means
Kimi K3.1 Pro is a large open-weight model with 2.8 trillion total parameters, 104 billion active parameters and a one-million-token context window, according to the SGLang cookbook. Alibaba Cloud describes it as a sparse mixture-of-experts model using KDA linear attention alongside DSA sparse attention, with multi-token prediction. The announcement notes that the model’s memory and high-speed scale-up interconnect needs make deployment a substantial infrastructure task. 3
17
In this context, day-zero adaptation means Alibaba Cloud says its M890 platform was prepared to run the newly released model weights promptly. Its claim of end-to-end validation covers two inference frameworks: vLLM and SGLang. That is a deployment-readiness milestone—not, by itself, a published benchmark or independent assessment of performance at scale. 3
Framework documentation shows configurations
SGLang’s cookbook lists PPU support for both M890 and 810E, and identifies a kimi-k3.1-day-0 branch based on main. 17 vLLM’s published recipe gives example configurations for Zhenwu PPUs: two 16-PPU 810E nodes using TP32/EP32, or four eight-PPU M890 nodes using TP8/DP4/EP32. The recipe says these configurations run the default MXFP4 checkpoint without speculative decoding.
25
27
Alibaba Cloud also published an M890P deployment guide specifying components including SGLang, SGLang Kernel, FlashMLA, Sailify and PLA, alongside SAIL-related configuration settings. Together, the cookbook, recipe and deployment guide make the support more concrete than an announcement alone: they document how the software stack is configured for these systems. 26
Joint adaptation and regional availability
Alibaba Cloud’s announcement attributes the adaptation to work across Alibaba Cloud, T-Head and Moonshot AI using the T-Head SAIL software stack. Related reporting names FlashMLA, DeepGEMM MoE, MegaMoE and MoonEP among the components or optimizations involved. These reports describe engineering work, but the available material does not provide independent measurements of its effects on speed, cost or reliability. 3
37
38
The November 5 announcement says Zhongwei opened invitation testing. M890 SuperPoD instances had previously launched in Ulanqab, where IT Home reported 64-card scale-up units and 800GB/s card-to-card interconnect bandwidth. The Ulanqab specifications provide context for the platform, but should not be treated as measured Kimi K3.1 Pro performance. 3
4
What the evidence does—and doesn’t—show
The clearest conclusion is that Alibaba Cloud announced day-zero Kimi K3.1 Pro adaptation on M890 and validation through vLLM and SGLang, while publishing documentation that specifies supported hardware and example configurations. 3
17
26
27
The material provided here does not verify the claimed Panmai 920 networking specifications or system-cost savings, customer adoption totals, or projected performance for a future Zhenwu V900. Nor do configuration recipes establish real-world throughput, latency or production reliability. Those claims require separate evidence.
Studio Global AI
Continue your research
This page includes a source-backed answer you can continue inside Studio Global.
People also ask
What is the short answer to "Alibaba Cloud’s M890 SuperPoD Adds Day-Zero Support for Kimi K3.1 Pro"?
Alibaba Cloud says its M890 SuperPoD completed day zero adaptation of Kimi K3.1 Pro weights and end to end validation with vLLM and SGLang; Zhongwei invitation testing also opened.