Kimi K3 is a multi-modal, open-weight foundation model that Moonshot AI says reaches frontier-level coding and general agentic performance. Bedrock already offered global and US-based inference for it, and India processing is now added for workloads with strict data residency requirements. Ultrahuman says it runs the model in production for its Jade health assistant, with health data staying inside AWS and Mumbai as a fallback Region.
Calls go through cross-Region inference profiles, and AWS handles the routing over its own network. The global profile, global.moonshotai.kimi-k3, is recommended for most uses, while the India profile limits processing to the Mumbai and Hyderabad Regions. The model supports Chat Completions, Responses, Converse and InvokeModel, with the OpenAI-compatible APIs recommended.
The post's example reads a scanned Hindi electricity bill and answers follow-up questions in Hinglish without a separate OCR step. It suggests setting the image detail parameter to high for small text and using explicit prompt caching to cut costs. Logs and metrics are stored in the Region where the request was submitted.
Why it matters if you build agents
Teams with India-based data residency rules can now call Kimi K3 without sending requests outside India. Developers should check the IAM permissions needed for the India profile, including access to both destination Regions. Pricing is per token, and the post recommends verifying extraction output before relying on it for important decisions.
Get stories like this every weekday
One short email with the day's agent tooling news.