Amazon Web Services has added Moonshot AI’s Kimi K3.1 Pro to Amazon Bedrock, its managed model marketplace, making the Chinese lab’s flagship model available to enterprise customers through AWS. The Highspeed variant of the model, which Moonshot says runs two to three times faster than the standard version, is the one on offer, and it is available in every region where Bedrock operates through cross-region inference. The move, announced on 6 November, follows Kimi K3’s arrival on the same platform in September.
What AWS announced
The announcement is short, and deliberately so. AWS says the model expands its open-weight portfolio and runs inside the same security boundary as proprietary models on Bedrock, with the governance controls customers already use. It is pitched at agentic work — coding and document processing — and AWS highlights high output tokens per second for faster, more responsive results.
AWS also notes the model supports explicit prompt caching, which cuts latency and input costs when the same context is reused across calls. What the announcement does not say is just as telling: no throughput figure for the Highspeed variant, no Bedrock pricing, and no benchmark scores. Those details will have to come from AWS documentation or independent testing.
The model behind the listing
Kimi K3.1 Pro is Moonshot’s flagship, launched on 23 October as the first model in its new K3.1 series. It is a post-trained version of Kimi K3, built with much larger reinforcement-learning compute across hundreds of downstream domains, and Moonshot calls it its most capable 3T-class open-weight model to date.
The headline numbers are a 2.8-trillion-parameter mixture-of-experts design with 104 billion active parameters per token, and a one-million-token context window. On Terminal-Bench 4.0, Moonshot’s own figures put it at 65.7 percent, ahead of GPT-6 Astra at 58.2 percent and just behind Claude Opus 5.5 at 66.4 percent.
The full model weights went public on Hugging Face on 4 November, as Moonshot had promised at launch, under its own Kimi K3.1 Pro licence. The Highspeed variant, though, arrived as a hosted-only speed tier: the same intelligence at two to three times the output speed, with no separate downloadable checkpoint.
A familiar route for Moonshot
Bedrock is not new territory for Moonshot. Kimi K3, the model K3.1 Pro builds on, arrived on the platform on 18 September, and AWS said at the time it was the first open-weight model on Bedrock to support explicit prompt caching. The K3.1 Pro listing is the next step in that sequence.
For Moonshot, which has no cloud of its own, Bedrock is the fastest route into Western enterprise infrastructure. The newer flagship follows the older model onto the same shelf, two weeks after its open weights went public, and the listing lands alongside a run of enterprise moves from the lab.
Our Read
Distribution is the story here, not capability. K3.1 Pro’s numbers were already public two weeks ago; what changed on 6 November is who can reach the model without touching Moonshot’s own infrastructure. Bedrock puts it behind AWS identity, billing, and compliance controls, which is exactly what large buyers ask for before they run a Chinese open-weight model in production.
The choice of the Highspeed variant is also deliberate. AWS is selling speed, not the downloadable weights: the fastest way to run the model is now a managed call, not a self-hosted deployment. That suits Moonshot too: the Highspeed tier never went on Hugging Face, and every Bedrock call is a customer it did not have to win itself.
The shelf it joins is crowded: OpenAI’s GPT-6.1 Terra arrived on Bedrock too, so K3.1 Pro competes on agentic coding scores rather than on novelty. The honest caveat is that none of those scores has been independently verified on Bedrock, and AWS published no speed or price figures to test the Highspeed claim against.
Why It Matters
For enterprises, the listing removes the biggest practical objection to using a Chinese open-weight model: running it outside their own security boundary. Bedrock customers can now call K3.1 Pro through the same console, billing, and guardrails as Anthropic and OpenAI models, with prompt caching to keep long-context costs down.
For Moonshot, it is the second enterprise door to open in two days, after Kimi Hosted Agents launched on 5 November. Together they show a lab moving from selling a model to selling managed ways to run it — and doing so through other companies’ platforms rather than building its own cloud.
Background
Moonshot has spent 2026 turning a beloved Chinese assistant into a model company: Kimi K3 shipped with open weights in July, K3.1 Pro followed on 23 October, and its weights went public on 4 November. The lab is also reported to be renting Blackwell GPUs in Europe while US export controls block shipments to China.
Source: AWS announcement and Moonshot’s launch post.
0 Comments