Executive Summary
Vercel has integrated Moonshot AI's Kimi K3 and Kimi K3 Fast models into its AI Gateway, making them accessible through US-based infrastructure providers like Baseten and Fireworks. This move is designed for teams with data residency and compliance requirements, offering features like automatic provider failover for high availability and Zero Data Retention (ZDR) for enhanced privacy. The integration allows developers to leverage the Kimi K3 models while ensuring their data is processed within the United States.
Key Takeaways
* New Models Available: Moonshot AI's `Kimi K3` and a premium, lower-latency version, `Kimi K3 Fast`, are now accessible.
* US-Based Providers: The models are served by US providers (initially Fireworks and Baseten), addressing data residency and compliance needs for customers.
* High Availability: AI Gateway automatically routes requests across multiple providers, ensuring failover, higher uptime, and greater throughput than a single provider.
* Zero Data Retention (ZDR): ZDR is supported for both models, allowing users to prevent their data from being stored post-inference.
* Pricing Tiers: `Kimi K3 Fast` costs approximately 50% more than the base model for lower latency. Specifying US-only inference adds a ~10% premium.
Strategic Importance
This integration makes a popular international AI model accessible and compliant for US enterprise customers, expanding Vercel's AI Gateway value proposition by addressing critical data security and residency concerns.