Kimi K3.1 Pro (Highspeed) now available on Amazon Bedrock: Moonshot's open-weight model with AWS guardrails

AWS announced on November 6, 2026 that Kimi K3.1 Pro from Moonshot AI is available on Amazon Bedrock. The listing covers the Highspeed variant, which AWS says delivers high output tokens per second for faster, more responsive results — aimed squarely at agentic use cases such as coding and document processing. According to Moonshot AI, Kimi K3.1 Pro is an open-weight, natively multimodal agentic model with a 1-million-token context window, and its most capable 3-trillion-parameter-class model to date.

It is a distribution story as much as a model story. Kimi K3.1 Pro launched on October 24, and its full weights landed on Hugging Face on November 4, so anyone can already self-host it. Bedrock is the managed alternative: no cluster to run, no weights to serve, and AWS's access controls, encryption, and auditing wrapped around the model instead.

What Bedrock buyers actually get

The AWS announcement is short on numbers and long on plumbing. Kimi K3.1 Pro runs within the same security boundary as Bedrock's proprietary models, and it supports explicit prompt caching, which reuses context across calls to reduce latency and input costs. It is available in all AWS Regions where Bedrock is available, through cross-Region inference — AWS routes requests to any supported commercial region. Developers start from the Bedrock console.

What AWS did not say is just as telling. The announcement gives no Bedrock pricing, no per-token rates or quotas, no latency or throughput figure for the Highspeed variant, and no benchmark results. Enterprises evaluating it against the self-hosted weights — or against Moonshot's own API — will have to wait for the pricing page to catch up, or test the model and measure it themselves.

The model behind the listing

Kimi K3.1 Pro is a post-trained version of Kimi K3, built on the same 2.8-trillion-parameter base with about 104 billion parameters active per token, and with post-training expanded across hundreds of downstream domains through reinforcement learning. Moonshot reported 65.7% on Terminal-Bench 4.0 at launch, close to the 66.4% it reported for Claude Opus 5.5 — the comparison that made the October launch notable. The Highspeed variant keeps the same intelligence and roughly doubles or triples output speed, according to Moonshot.

One deliberate limit travels with the model: Moonshot scrubbed cyber-vulnerability data from K3.1 Pro's training, cutting its exploit success rate by about 9 points before the weights went public. That is the same instinct behind the Bedrock listing — the model arrives with guardrails attached, whether the guardrails are baked into the weights or supplied by AWS's managed runtime.

A crowded open-weight shelf

Bedrock's open-weight shelf has filled fast. Kimi K3 itself arrived on Bedrock on September 18, the first time a Chinese open model entered a global cloud provider's catalog on a revenue-sharing basis. Since then AWS has added Z.ai's GLM 5.3, xAI's Grok 4.7, and OpenAI's GPT-6.1 Terra. Kimi K3.1 Pro is the newest name on that list, and the second Moonshot model after K3.

The pattern is deliberate. On Amazon's third-quarter earnings call, CEO Andy Jassy said the Western world is "very hungry" for homegrown open models — and AWS is adding the Chinese open-weight systems anyway, because enterprise buyers want the choice. Bedrock's pitch is that the model's origin matters less than the runtime around it: same security boundary, same governance, same console, whichever lab built the weights.

What to watch

Three things decide whether this listing matters beyond the announcement. First, Bedrock pricing: until AWS publishes it, the managed route cannot be compared with Moonshot's own API or with self-hosting the public weights. Second, real throughput: "high output tokens per second" is a claim, not a number, and the Highspeed variant's 2–3x speedup comes from Moonshot's own materials, not an AWS measurement. Third, how Moonshot's own hosted agent service, launched the day before, and the Bedrock listing divide the market — one sells Moonshot's harness, the other sells AWS's runtime, and both sell the same model.

For now, the news is simple: one of the strongest open-weight models of the autumn is a Bedrock model too, available today, in every region Bedrock reaches.