Introducing Kimi K3 on Amazon Bedrock
Amazon Web Services Alex Thewsey
AWS added Kimi K3 to Bedrock, bringing Moonshot AI’s huge open model to coding and knowledge work. It also gets prompt caching and Bedrock’s security controls, which is the real draw.
Based on reporting by Amazon Web Services, Alex Thewsey — read the original for the full story.
Summary, retelling and take written by AI under human oversight; images are AI-generated illustrations. How we work · Report an error
AWS has put Kimi K3 from Moonshot AI onto Amazon Bedrock, giving developers another open-weight model to use for coding and knowledge tasks. Moonshot says Kimi K3 is its most capable model and the first open model to reach 2.8 trillion parameters. It also has native vision, a 1-million-token context window, and about 2.5x better scaling efficiency than Kimi K2.
The practical pitch is simple: this is a model built for work that goes on for a while. Long codebases, dense documents, and image-heavy workflows need a model that can keep context without falling apart halfway through the job. Kimi K3 is also the first open-weight model on Bedrock with explicit prompt caching, so repeated context can cut both latency and input costs.
AWS is leaning hard into open-weight support on Bedrock. Since 2025, the service has added dozens of models from providers including DeepSeek, Google, MiniMax, Mistral AI, Moonshot AI, NVIDIA, OpenAI, and Qwen. In 2026, Bedrock added support for tool calling, structured output, reasoning, response streaming, and the Responses and Chat Completions APIs, and those platform features can apply to new models as they arrive.
AWS is also making a clear security pitch. With Bedrock, data stays inside the AWS data boundary, is not shared with the model provider, and is not used to train the underlying model. Zero data retention is enabled for inference requests, and zero operator access keeps AWS operators out of prompts and completions. Kimi K3 is available through the US Geo and Global cross-Region inference profiles, with global cross-Region inference costing about 10% less than a geographic profile.
My take — AI-written commentary, not fact-checked reporting
This is the kind of open-model story that actually matters: less hype, more plumbing. The interesting part isn’t that Kimi K3 is enormous; it’s that AWS is wrapping giant open weights in caching, APIs, and controls people can ship with. That’s how open models stop being demos and start becoming boring infrastructure, which is the compliment they should want.
Read more about this at: Amazon Web Services
