Report: Microsoft tests Kimi K3 to replace GPT and Claude in Copilot
Redmond pegs the swap at up to $600M in saved inference, a model-layer commoditization signal, even as Washington weighs banning Chinese open weights.
Copy markdown
Kimi K3 heads to Azure — and maybe Copilot
Microsoft has reportedly started adding Moonshot's open-weight Kimi K3 to Azure and is testing whether it can handle Copilot tasks currently run on GPT-5.6 and Claude. For builders, that means a frontier-class open coder landing in the Azure catalog you already deploy from.
A $600M bet the model layer is commoditizing
Microsoft reportedly calculates up to $600M in inference savings from swapping closed frontier models for Kimi K3 — a loud signal that open weights are now close enough to GPT/Claude for production, and downward pressure on what that tier costs you.
Washington could still pull the plug
The Trump administration is reportedly reviving a push to ban Chinese open-source models after Kimi K3's launch, weighing an executive order and Entity List additions. Reporters note downloadable weights make an outright ban near-impossible to enforce, but US teams self-hosting Kimi, DeepSeek, or Qwen should watch this closely.