Moonshot AI's Kimi K3.1 Pro Highspeed lands on Amazon Bedrock with 1M-token context

Kimi K3.1 Pro Highspeed from Moonshot AI is now available on Amazon Bedrock, with a 1-million-token context window and the same endpoint, APIs, and features as Kimi K3 on the platform, according to an AWS Machine Learning blog post.

Model details

Kimi K3.1 Pro Highspeed is part of Moonshot AI's new K3.1 series. It is a post-trained version of Kimi K3 with the same 3T-class open-weight architecture, trained with substantially more post-training compute. According to Moonshot AI, that extra training targets long-horizon agentic work and fewer factual errors. Moonshot AI also says it improves professional tasks such as financial modeling and office documents, and coding for game engines, 3D scenes, and CAD scripts.

The Highspeed variant generates output tokens at a higher rate. Moonshot AI reports 2-3 times the speed of Kimi K3.1 Pro. Supported features are tool calling, structured output, configurable reasoning, image input, and streaming. Per-token Standard pricing is not yet disclosed.

Access on Bedrock

The model is available on the bedrock-runtime endpoint through the OpenAI-compatible Chat Completions and Responses APIs, and through the Bedrock Converse and Invoke APIs. AWS recommends Chat Completions. Teams already using Kimi K3 can switch by changing the model ID.

Access runs through cross-Region inference profiles:

• global.moonshotai.kimi-k3.1: routes to any supported commercial AWS Region and costs less per token than the geographic profiles. AWS recommends it for most workloads.
• us.moonshotai.kimi-k3.1: keeps processing within US Regions.
• in.moonshotai.kimi-k3.1: keeps processing within India Regions.

The model always reasons before answering. The reasoning_effort parameter (or reasoning.effort in the Responses API) accepts low, high, or max. If unset, the model uses max, so AWS notes that a lower value is the most direct way to cut latency and output tokens on simpler tasks.

Cost controls

The Flex tier serves the model at half the Standard price, in exchange for potentially longer response times. AWS positions it for model evaluations, content summarization, and background agent tasks. Tiers are chosen per request, so one application can mix Priority, Standard, and Flex. On-demand quota is shared across all three. Flex is set with service_tier on Chat Completions, or serviceTier on Converse. CloudWatch exposes ServiceTier and ResolvedServiceTier dimensions.

The model caches repeated prefixes automatically. Explicit prompt caching lets developers mark where the stable prefix ends. The rules:

• It works on Chat Completions, Responses, and Invoke, but not Converse.
• A prompt_cache_breakpoint must follow at least 256 tokens of stable content.
• Cached content is retained for at least 30 minutes.
• Cache writes are billed above standard input. Cache reads are billed at a discount and do not count toward the input tokens-per-minute quota.

Data handling

AWS states that 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, and zero operator access means AWS operators cannot view prompts and completions during inference.

The post also walks through using the model as a coding assistant in OpenCode.

What this means

For teams already running Kimi K3 on Bedrock, adoption is a model ID change, with no new endpoint or API integration. The default reasoning setting of max means latency-sensitive workloads should set reasoning_effort explicitly. The Flex tier at half the Standard price gives a defined option for non-urgent work, and the shared quota across tiers means traffic can be split without separate capacity planning. The speed and accuracy gains are Moonshot AI's own claims, so teams should test them on their own workloads before moving production traffic.