Amazon Web Services said Kimi K3.1 Pro Highspeed, from Moonshot AI, is now available on Amazon Bedrock, expanding the cloud’s open-weight model lineup with a 1-million-token context window and faster output, according to an AWS machine-learning blog and a what’s-new notice.
The model is a post-trained version of Kimi K3 with the same 3-trillion-parameter-class open-weight architecture, AWS said, citing Moonshot AI’s claim that extra post-training targets long-horizon agentic work, fewer factual errors, and better performance on financial modeling, office documents, game-engine and CAD scripting, and 3D scenes. Moonshot AI reports that Highspeed generates output tokens at two to three times the speed of Kimi K3.1 Pro, AWS wrote. Those speed and quality figures are the vendor’s claims. The model supports tool calling, structured output, configurable reasoning, image input and streaming, and can be reached by changing the model ID in existing Kimi K3 code, the blog said.
AWS said customer data is processed within the AWS data boundary, is not shared with Moonshot AI, is not used to train the model, and that zero data retention is always enabled for inference, with no operator access to prompts and completions. A Flex tier serves the model at half the standard price in exchange for potentially longer response times. Cross-Region inference profiles include a lower-cost global route and U.S. and India residency options. AWS recommended the OpenAI-compatible Chat Completions API and noted that if reasoning effort is left unset, the model uses “max.”