Amazon Bedrock Adds In-Country Inference in India for Moonshot AI Kimi K3

AWS has enabled in-country inference within India for Moonshot AI's multi-modal, open-weight foundation model Kimi K3 on Amazon Bedrock. The feature uses geographic cross-Region inference profiles (`in.moonshotai.kimi-k3`) to restrict model processing to the Mumbai and Hyderabad AWS Regions, serving organizations with strict domestic data residency requirements. Ultrahuman reported using Kimi K3 on Bedrock with prompt caching to reduce LLM costs by 80 percent and accelerate multi-turn chats by 85 percent for its health assistant Jade across over 14,000 daily conversations. AWS also demonstrated an end-to-end pipeline leveraging the model's direct multimodal reading capabilities to extract structured JSON data from Indic-language documents without dedicated OCR services.

Key takeaways

• Amazon Bedrock launched the `in.moonshotai.kimi-k3` inference profile, ensuring data processing remains within AWS Mumbai and Hyderabad Regions.
• Health-tech firm Ultrahuman reported an 80 percent reduction in LLM costs and an 85 percent speedup in multi-turn chats using Kimi K3 with prompt caching on Bedrock.
• The model directly processes scanned documents and images to output structured JSON and support multilingual question-answering without separate OCR tools.