Microsoft is reportedly preparing to add Moonshot AI’s Kimi K3 model to Azure while Copilot engineers evaluate whether it can handle selected workloads now served by OpenAI and Anthropic systems. However, Microsoft has not announced a deployment.
TL;DR
- Reports say Microsoft is preparing Kimi K3 for Azure and testing it for possible Copilot use.
- The reported evaluation could lower inference costs, but the widely cited $600 million estimate remains unconfirmed.
- Kimi K3 follows Kimi K2.7 Code, which GitHub already offers in Copilot and hosts on Microsoft Azure.
Microsoft is evaluating Moonshot AI’s Kimi K3 for potential use inside Copilot, according to a report from The Information cited by Crypto Briefing. The company is also said to be working on adding the model to its Azure cloud service.
The report claims a successful shift of some Copilot workloads could reduce Microsoft’s annual AI inference costs by as much as $600 million. Microsoft has not publicly confirmed that figure, identified which Copilot features are being tested, or announced when Kimi K3 could become available through Azure.
This distinction matters. An internal evaluation does not mean Microsoft has adopted Kimi K3 or deployed it to users. Engineers would still need to assess response quality, reliability, safety, latency, infrastructure requirements, and compliance before the model could support a production service.
Microsoft’s Multi-Model Copilot Strategy
The reported test fits Microsoft’s broader effort to reduce dependence on a single model provider. In March, Microsoft described Microsoft 365 Copilot as “model diverse by design,” adding that customers should receive greater choice, performance, and flexibility across an open environment.
Microsoft 365 Copilot currently uses models from OpenAI and Anthropic. GitHub, which Microsoft owns, has already expanded that mix by making Moonshot AI’s Kimi K2.7 Code generally available in GitHub Copilot.
GitHub called Kimi K2.7 Code the first open-weight model available through the Copilot model picker and said it offers users “more choice and a lower-cost option” for coding workflows. The model is hosted by GitHub on Microsoft Azure, creating a clear technical and commercial precedent for evaluating Kimi K3.
GitHub also requires Business and Enterprise administrators to enable Kimi K2.7 Code and recommends reviewing open-weight models against security, compliance, and data-governance requirements.
Topics For More Insights
What Kimi K3 Brings
Moonshot AI introduced Kimi K3 on July 16 as its “most capable model.” The company says it has 2.8 trillion parameters, native vision capabilities, and a one-million-token context window for long-horizon coding, knowledge work, and reasoning.
Kimi K3 uses a Mixture of Experts design that activates 16 of 896 experts for each token. Moonshot said the model still trails the strongest proprietary systems overall, while claiming frontier-level results across its evaluation suite.
The company plans to release the full model weights by July 27, 2026. Until those weights are available and independently tested, many of Moonshot’s performance and efficiency claims remain provisional.
What Happens Next
Microsoft has not issued an official announcement confirming Kimi K3 for Azure or Copilot. The immediate development is therefore a reported evaluation, supported by Microsoft’s existing use of Kimi K2.7 Code and its public commitment to model diversity.
A confirmed rollout would give Microsoft another open-weight option for routing selected AI tasks. For now, the more accurate conclusion is that Microsoft is testing Kimi K3, not that it has adopted the model across Copilot.
