Kimi K3.1 Pro Highspeed is on Amazon Bedrock at $4 in, $20 out

Moonshot AI's Kimi K3.1 Pro Highspeed is available on Amazon Bedrock, AWS said in a What's New notice dated 6 November 2026 and a longer blog post dated 10 November. Bedrock's pricing page lists it at $4.00 per million input tokens and $20.00 per million output tokens on the global inference profile, on the Standard tier.

K3.1 Pro is a post-trained version of Kimi K3 with the same architecture and the same 1-million-token context window. Moonshot AI's model card on Hugging Face gives it 2.8 trillion total parameters, 104 billion of them activated, under Moonshot's own licence. According to Moonshot AI, as relayed by AWS, the extra training targets long-running agent work and fewer factual errors, and the Highspeed variant generates output at two to three times the speed of K3.1 Pro. AWS publishes no benchmark and no tokens-per-second figure to back either claim.

What it costs

Prices below are from the Bedrock pricing page, per million tokens, Standard tier.

• Global profile: $4.00 input, $20.00 output, $0.20 cache read, $5.00 cache write (30 minutes).
• US and India profiles, which keep processing in those regions: $4.40 input, $22.00 output, $0.22 cache read, $5.50 cache write.
• Kimi K3 on the global profile, for comparison: $3.00 input, $15.00 output, $0.30 cache read, $3.75 cache write.

So the new model costs a third more than K3 on input and output, and a third less on cache reads. Moonshot AI's own price list shows the same $4 and $20 for the Highspeed variant, and a standard-speed kimi-k3.1-pro at $2 and $10. The AWS post covers only Highspeed; it does not say whether the cheaper standard variant will come to Bedrock.

AWS says the Flex tier serves the model at half the Standard price in exchange for potentially longer response times, and suggests it for evaluations, summarisation and background agent tasks. The tier is chosen per request, and quota is shared across Standard, Priority and Flex.

What to change in your code

• Model ID: global.moonshotai.kimi-k3.1, or us.moonshotai.kimi-k3.1 and in.moonshotai.kimi-k3.1 for data residency. AWS says teams already on Kimi K3 change only the model ID.
• Reasoning: the model always reasons before it answers and defaults to max effort. The levels are low, high and max.
• Caching: repeated prefixes are cached automatically. Explicit breakpoints need at least 256 tokens of stable content before them, cached content is kept for at least 30 minutes, and cache reads do not count toward the input tokens-per-minute quota.
• APIs: OpenAI-compatible Chat Completions and Responses, plus Bedrock's Converse and Invoke. Explicit caching is not available on Converse.

The model supports tool calling, structured output, image input and streaming. AWS says prompts and completions stay inside the AWS data boundary, are not shared with Moonshot AI and are not used for training, and that zero data retention is always on for inference requests.

What the announcement leaves out

There are three gaps. The post names US and India geographic profiles and no European one; the notice says the model is reachable from every Bedrock region through cross-region inference, but a team that must keep processing inside the EU has no geographic profile named for it. The pricing page shows Standard-tier prices only, so the Priority price is not stated. And the speed claim is Moonshot AI's, without a number from AWS.

A max-effort default on a $20-per-million-output model is the line to watch. Before moving agent traffic over, run your own task set at low and high effort, record output tokens per task, and compare cost per completed task against Kimi K3, not price per token.