Moonshot AI’s Kimi K3.1 Pro Highspeed arrives on Amazon Bedrock

A faster model for long-horizon work

Kimi K3.1 Pro Highspeed is part of Moonshot AI’s K3.1 series and uses the same 3T-class open-weight architecture as Kimi K3. According to AWS, Moonshot AI trained the model with substantially more post-training compute.

Moonshot AI says the additional training targets long-horizon agentic work and fewer factual errors. The company also reports improvements in professional tasks including financial modeling and office documents, as well as coding for game engines, 3D scenes and computer-aided design scripts.

The Highspeed model generates output tokens at a higher rate. Moonshot AI reports that it is 2, 3 times the speed of Kimi K3.1 Pro.

Existing Bedrock applications can switch model IDs

Developers already building with Kimi K3 can try Kimi K3.1 Pro Highspeed by changing the model ID in existing code. The model supports tool calling, structured output, configurable reasoning, image input and streaming.

AWS provides access through the bedrock-runtime endpoint and supports the OpenAI-compatible Chat Completions and Responses APIs, along with the Amazon Bedrock Converse and Invoke APIs. AWS recommends the Chat Completions API.

Requests use cross-Region inference profiles. global.moonshotai.kimi-k3.1 can route requests to any supported commercial AWS Region and costs less per token than geographic profiles, according to AWS. The us.moonshotai.kimi-k3.1 profile keeps processing within US Regions, while in.moonshotai.kimi-k3.1 keeps it within India Regions for workloads with data residency requirements.

Kimi K3.1 Pro Highspeed always reasons before answering. With the Chat Completions API, developers can set reasoning_effort to low, high or max. AWS says using a lower value can reduce latency and output tokens on simpler tasks.

Flex pricing and prompt caching target operating costs

Amazon Bedrock’s Flex tier serves Kimi K3.1 Pro Highspeed at half the Standard price in exchange for potentially longer response times. AWS positions it for model evaluations, content summarization and background tasks in agentic workflows, where cost efficiency has a higher priority than response time.

Applications can select a tier for each request, allowing the same application to use Priority for latency-critical requests, Standard for everyday traffic and Flex for cost-sensitive work. AWS recommends moving a small share of traffic to Flex first and comparing latency and cost.

The model also automatically caches repeated prefixes. Developers can use explicit prompt caching to mark stable content such as system instructions, tool definitions, repository conventions and reference documents. AWS says this can reduce latency and input cost when long context is reused. Explicit caching is supported on the Chat Completions, Responses and Invoke APIs, but not on Converse. A prompt cache breakpoint must follow at least 256 tokens of stable content, and cached content is retained for at least 30 minutes.

Security controls and coding workflows

AWS says data is processed within the AWS data boundary, is not shared with Moonshot AI and is not used to train the model. Zero data retention is always enabled for inference requests, while zero operator access means AWS operators cannot view prompts and completions during inference.

AWS also demonstrates Kimi K3.1 Pro Highspeed as a coding assistant in OpenCode, an open source, model-agnostic coding agent that runs in a terminal. Developers can connect it through OpenCode’s OpenAI-compatible provider and the Amazon Bedrock Chat Completions API.

For users, the release offers a path to adopt the model without changing an existing Kimi K3 integration or security posture. For teams managing repeated context or background workloads, the Flex tier and prompt caching provide AWS-supported ways to trade response time and caching costs against inference expense.