Dispatch
Vercel AI Gateway adds Kimi K3 with US-based providers and Zero Data Retention
Vercel's July 27 changelog makes Kimi K3 and the faster-serving Kimi K3 Fast available on AI Gateway from US-hosted providers including Baseten and Fireworks, which matters for teams that want the leading open-weights model without routing prompts to China-based inference. Zero Data Retention is supported and can be switched on globally in the Gateway dashboard or per-request via the zeroDataRetention parameter. Vercel does not publish flat pricing — it varies by provider and variant, with K3 Fast running ~10% above the base model and regional variants also ~10% higher, so the model endpoints API is the only reliable way to compare.
↳ Follow the thread