Kimi K3 and Kimi K3 Fast with ZDR and US-based providers now on AI Gateway
By Steven Van ·
Both Kimi K3 models run on US providers with zero data retention, with fast mode about 50% dearer and US-only inference about 10% dearer.
Kimi K3 and Kimi K3 Fast can now be served from US-based providers on Vercel AI Gateway, including Baseten and Fireworks. Zero Data Retention (ZDR) is also supported for both models, as Vercel announced on 27 July 2026. Teams with data residency or compliance requirements can use the model on US infrastructure.
Routing and failover
AI Gateway serves Kimi K3 from multiple providers and routes across them automatically for failover, higher uptime and more available throughput than any single provider offers. The model ID stays moonshotai/kimi-k3, and the gateway handles provider selection and fallback.
Kimi K3 Fast
Kimi K3 Fast is a faster serving path that trades a higher per-token cost for lower latency. It costs about 50% more than the base model. There are two ways to request it:
- Set the speed option to fast under providerOptions.gateway on the base model. The model stays moonshotai/kimi-k3, and requests fall back to standard speed when the fast tier is unavailable.
- Use the model ID moonshotai/kimi-k3-fast directly.
US inference
To keep Kimi K3 requests on US data centers only, set inferenceRegion to { scope: 'zone', geoRegion: 'us' } under providerOptions.gateway. Regional pricing is about 10% more than the regular variant.
Zero Data Retention
ZDR can be turned on for every request in the AI Gateway dashboard settings, or per request by setting zeroDataRetention: true under providerOptions.gateway.