Beijing-based Moonshot AI has cemented its position at the center of the global open-weight model ecosystem with the release of Kimi K3.1 Pro, a massive 2.8-trillion-parameter Mixture-of-Experts (MoE) foundation model engineered for multi-step software engineering, deep financial research, and autonomous tool execution.
The launch arrives alongside a sobering benchmark for Western AI labs: according to a November 2026 enterprise survey published by Silicon Valley venture capital firm Menlo Ventures, Chinese AI models now capture 89 percent of all global enterprise spending on open-source and open-weight models. Moonshot AI’s Kimi family alone commands 61 percent of global open-source enterprise spend, outpacing Meta’s Llama, Mistral, and Western open-weight rivals.
Inside Kimi K3.1 Pro: 2.8 trillion parameters and 1M context
Kimi K3.1 Pro scales Moonshot’s sparse MoE routing architecture to 2.8 trillion total parameters while keeping active parameter overhead tightly constrained per token, enabling high-throughput inference across a 1-million-token context window. Within hours of Moonshot publishing the open weights on Hugging Face, major US cloud providers—including Amazon Bedrock, Microsoft Foundry, and Nvidia NIM—added managed API endpoints for Kimi K3.1 Pro.
Why global enterprises are standardizing on Chinese open weights
Chief information officers cited three factors driving Kimi K3.1 Pro adoption: permissive commercial licensing that allows on-premise fine-tuning without data leakage, aggressive FP8 quantization support on both Nvidia Blackwell and domestic clusters, and superior multi-file repository reasoning across 500,000+ token codebases.
- Chinese open-weight AI models now account for 89% of global enterprise spending on open-source AI according to Silicon Valley venture firm Menlo Ventures, with Moonshot AI’s Kimi capturing 61% market share.
- Moonshot AI has launched Kimi K3.1 Pro, a 2.8-trillion-parameter Mixture-of-Experts (MoE) model with a 1-million-token context window engineered for long-horizon agentic coding and quantitative financial research.
- Kimi K3.1 Pro weights are available on Hugging Face alongside managed enterprise deployments on Amazon Bedrock, Microsoft Foundry, and Nvidia NIM microservices.
“While US frontier labs locked their best reasoning checkpoints behind closed APIs, Chinese labs led by Moonshot AI and DeepSeek turned open-weight MoE architectures into the default infrastructure layer for global enterprise developers.”— Menlo Ventures Enterprise Open-Source AI Report, November 2026
- ✓Sources: written from primary sources and official statements, linked above.
- ✓Fact-check: names, dates and claims checked against the original sources.
- ✓AI use: researched with AI tools; written and reviewed by our editors. Our AI policy
- ✓Independence: UpdateNext has no financial relationship with the companies covered.



