Kimi K3.1 Pro vs Qwen3.8-Max: which giant open model fits you?

Kimi K3.1 Pro and Qwen3.8-Max are the largest open-weights models you can download. Compare size, input types, memory needs and licenses side by side.

  • Updated

Verdict. Both are trillion-parameter mixture-of-experts models with a 1M-token context, and both need a multi-GPU data-center setup, so most people will reach them through an API. Choose Kimi K3.1 Pro for the newer release with the smaller download, built for long coding and office tasks. Choose Qwen3.8-Max if you need video input. Neither developer's benchmark numbers are independent, so test both on your own tasks.

Specifications

Specifications Kimi K3.1 ProQwen3.8-Max
DeveloperMoonshot AIAlibaba Qwen
ReleasedNovember 2, 2026August 12, 2026
Total parameters2.8T (K3.1 Pro)2.45T (2.4T-A95B)
Active parameters104B95B
ArchitectureMixture of expertsMixture of experts
Context window1M tokens1M tokens
InputText, ImageText, Image, Video
License Kimi K3 License (Conditional) Qwen3.8-Max License (Conditional)
Memory for the smallest variant (4-bit)1.72 TB (K3.1 Pro)1.69 TB (2.4T-A95B)

Performance claims come from the developer and have not been independently verified.

Pick Kimi K3.1 Pro if

  • You want the newer model: it was published on 2 November 2026, against 12 August for Qwen3.8-Max.
  • You want the smaller download: about 1.56 TB of weights, against about 2.5 TB for Qwen's FP8 checkpoint.
  • Your work is long coding or office tasks, which is what Moonshot built this version for.
  • You are not a hosting business with over US$20 million of combined revenue, which would need a separate agreement.

Pick Qwen3.8-Max if

  • You need video understanding as well as text and images.
  • You work with long financial reports and PDFs, which Alibaba highlights for documents of more than 200 pages.
  • You do not run a hosting or AI work-assistant business with over US$50 million of combined revenue, which would need a separate license from Qwen.
  • You want a smaller sibling for development: Qwen3.8-27B runs on a single GPU.

Kimi K3.1 Pro · Qwen3.8-Max