Kimi K3.1 Pro Highspeed from Moonshot AI is now available on Amazon Bedrock, bringing a 3T-class open-weight architecture with a 1-million-token context window to AWS infrastructure, aws.amazon.com reported on November 11, 2026.
Kimi K3.1 Pro Highspeed improves performance and output speeds
The post-trained model features the same underlying architecture as Kimi K3 but incorporates substantially more post-training compute. Moonshot AI directed this extra training toward long-horizon agentic work and the reduction of factual errors.
Professional tasks also receive targeted improvements. These cover financial modeling, office documents, coding for game engines, 3D scenes, and computer-aided design scripts.
Kimi K3.1 Pro Highspeed generates output tokens at a higher rate. Moonshot AI reports that output speeds reach two to three times the rate of the standard Kimi K3.1 Pro.
The model shares the exact endpoint, APIs, and feature set as Kimi K3 on Amazon Bedrock. Developers already building with Kimi K3 can adopt the new version simply by changing the model ID in their existing code.
🚀 Moonshot AI’s Kimi K3.1 Pro (Highspeed) lands on Amazon Bedrock – an open‑weight, multimodal agentic model with a 1‑million‑token context window and 3‑trillion‑parameter class power. Built for coding, docs, with prompt caching & cross‑region inference. #AI #AWS
Developers access features through various APIs and routing profiles
Supported features include tool calling, structured output, configurable reasoning, image input, and streaming. Developers can access the model on the bedrock-runtime endpoint through the OpenAI-compatible Chat Completions and Responses APIs, as well as the Amazon Bedrock Converse and Invoke APIs.
Amazon Bedrock routes requests through cross-Region inference profiles. The global profile, global.moonshotai.kimi-k3.1, routes requests to any supported commercial AWS Region at a lower cost per token than geographic profiles, making it the recommended choice for most workloads.
Workloads with strict data residency requirements can use geographic profiles instead. The US profile keeps processing within US Regions, while the India profile restricts processing to India Regions.
Flex tier reduces costs for background tasks
The Flex tier serves Kimi K3.1 Pro Highspeed at half the Standard price in exchange for potentially longer response times. This tier fits background tasks in agentic workflows, model evaluations, and content summarization.
Applications can mix pricing tiers within the same deployment. Priority handles latency-critical requests, Standard manages everyday traffic, and Flex takes on cost-sensitive work while sharing the same on-demand quota.
Developers set the tier per request by designating the service tier in the API call. Amazon CloudWatch tracks traffic distribution across tiers using specific service tier dimensions.
Prompt Caching and Token Efficiency
Agents and coding assistants repeatedly send large prefixes containing system instructions, tool definitions, and reference documents. Kimi K3.1 Pro Highspeed caches these repeated prefixes automatically.
Explicit prompt caching allows developers to mark where stable prefixes end to improve cache hit rates. This mechanism reduces latency and input costs for long, repeated context across turns.
Explicit caching applies to the Chat Completions, Responses, and Invoke APIs, but excludes the Converse API. Cache breakpoints require at least 256 tokens of stable content, and cached content remains available for a minimum of 30 minutes.
Terminal Integration with OpenCode
OpenCode runs as an open-source, model-agnostic coding agent directly inside the terminal. Developers connect the tool to Kimi K3.1 Pro Highspeed using the Amazon Bedrock Chat Completions API and an API key.
Output speed dictates performance in agentic coding loops. Because coding agents execute sequential steps—reading files, planning changes, editing code, and running tests—faster token generation shortens the entire edit-test-fix cycle.
