Kimi K3.1 Pro on Amazon Bedrock — No Price, No Speed Number Yet

Amazon Web Services announced on 6 November 2026 that Kimi K3.1 Pro from Moonshot AI is now available on Amazon Bedrock, with a 1-million-token context window and a faster Highspeed variant for coding and document work. The catch: no Bedrock price, no throughput number, and no benchmark score of any kind. Amazon Bedrock is AWS's managed service for running foundation models through its APIs, so teams do not have to host the model weights themselves. Kimi K3.1 Pro (Highspeed) is Moonshot AI's faster version of its flagship model, built for agentic work where the model runs long, multi-step jobs on its own.

What AWS actually announced

The announcement is short, and its wording matters. AWS says Kimi K3.1 Pro is available on Bedrock today, expanding its open-weight model portfolio. Customers can now run agentic use cases such as coding and document processing on Kimi K3.1 Pro (Highspeed), which AWS says delivers high output tokens per second, known as OTPS, for faster and more responsive results. That is the full speed claim. AWS gives no number for the output rate, no latency figure, and no comparison against the standard variant or any rival model.

According to Moonshot AI, as relayed by AWS, 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. Moonshot reports improvements in coding, 3D creation, financial research, and other agentic work. On Bedrock, the model runs within the same security boundary as proprietary models, and it supports explicit prompt caching, which reuses repeated prompt text to reduce latency and input costs. The model is available in all AWS Regions where Amazon Bedrock is available, through cross-Region inference, which routes requests across regions rather than tying the model to one region. To start, users open the Amazon Bedrock console and pick the model in the playground.

That availability detail is the real news for teams already on AWS. Before this announcement, the main managed route was Moonshot's own API and apps. Now an AWS team can call the model through Bedrock's APIs, inside the same account, billing, and access controls. AWS says inference data for open-weight models stays within the AWS data boundary, is not shared with the provider, and is not used to train the model — the same promise it made when Kimi K3 arrived on Bedrock on 18 September 2026.

Illustration created for INORYVA

What AWS did not say

None of the following appears in the AWS announcement, and this article does not fill the gaps with numbers from other sources presented as Bedrock facts.

Detail	What AWS published
Bedrock price per million tokens	Not published in the announcement
Highspeed throughput (tokens per second)	Not published — only "high OTPS"
Benchmark scores on Bedrock	None published
Quotas or rate limits	Not published in the announcement
Context window	1 million tokens, per Moonshot AI
Model scale	3-trillion-parameter class, per Moonshot AI
Availability	All Bedrock Regions, via cross-Region inference
Prompt caching	Explicit prompt caching supported

The missing price is the biggest gap for anyone planning a budget. AWS normally lists Bedrock prices on its pricing page, but the announcement states no rate, and this article will not quote a third-party tracker as if it were the Bedrock price. Moonshot's own API prices are Moonshot's, not AWS's. Until AWS publishes the Bedrock rate, the honest statement is that it is not published.

The missing speed number matters just as much. Highspeed is the whole point of this variant, yet AWS gives no tokens-per-second figure, no latency range, and no test conditions. A team choosing between the standard and Highspeed variants on Bedrock therefore cannot judge the speed gain from the announcement alone. The only practical check is to run the model in the Bedrock playground and measure it on real work.

The model behind the announcement

Kimi K3.1 Pro itself is not new. Moonshot AI launched it on 23 October 2026, about two weeks before the Bedrock announcement, as the first model in its K3.1 series. It is a post-trained version of Kimi K3, put through far more reinforcement learning across what Moonshot calls hundreds of downstream domains. Moonshot's model card puts it at 2.8 trillion parameters in total, with 104 billion active for each token, and a 1-million-token context window. It accepts text and image input.

Moonshot's own launch benchmarks put Kimi K3.1 Pro close to the best closed models on agentic coding. On Terminal-Bench 4.0, which grades coding agents on realistic terminal tasks, Moonshot reports 65.7 percent, against 66.4 percent for Anthropic's Claude Opus 5.5 and 58.2 percent for OpenAI's GPT-6 Astra — a 0.7-point gap to the leader, on Moonshot's own table. Moonshot also reports 51.7 percent on SWE-Marathon, ahead of Opus 5.5, and a hallucination rate on its own hardest factual test down from 23.4 percent to 8.3 percent. All of these figures are vendor-reported, not independently checked on Bedrock, and AWS repeated none of them. Treat them as Moonshot's claims, not AWS's measurements.

One wrinkle: Moonshot promised the full weights by 4 November 2026, and reports that day said they had landed on Hugging Face. But Bedrock is an API route, not a download, so teams wanting the weights should check Hugging Face directly.

Illustration created for INORYVA

How it compares with the flagship field

The Bedrock arrival puts Kimi K3.1 Pro alongside the other flagships AWS hosts. Our flagship AI model comparison tracks Gemini Carbon, Claude Opus 5.5, and GPT-6 Astra, where the three leaders sit within about a point of each other. Launch-day reporting placed Kimi K3.1 Pro fourth on that index, but that came from launch coverage, not a Bedrock measurement, and it is not in the AWS announcement.

Against its closest relatives, the story is clearer. Kimi K3.1 Pro is the post-trained successor to Kimi K3, on Bedrock since September. The upgrade is about agentic stamina — longer runs, better tool use, fewer hallucinations — according to Moonshot. GPT-6.1 Terra reached Bedrock on its launch day with pricing stated upfront, which is the contrast that makes this announcement feel thin. Mistral Large 4 is the other open-weight flagship in the same conversation, and its weights are downloadable, a different kind of openness from a Bedrock API listing.

The best first users are AWS teams already running long agent jobs: try it in the Bedrock playground on a real task, measure tokens per second and cost per task, and compare it against whatever you run today before moving production traffic.

Illustration created for INORYVA

What to watch next

Three things will complete this story: a Bedrock per-token rate, an independent throughput measurement for Highspeed, and an independent benchmark run through the Bedrock endpoint. Until those arrive, the honest summary is short: Kimi K3.1 Pro is on Bedrock, it has a 1-million-token context, and AWS has not yet said what it costs or how fast it runs there.

Key Takeaways

• AWS announced on 6 November 2026 that Kimi K3.1 Pro (Highspeed) is available on Amazon Bedrock in all Bedrock Regions through cross-Region inference.
• The model is Moonshot AI's open-weight flagship, with a 1-million-token context window, per Moonshot.
• AWS published no Bedrock price, no throughput figure for the Highspeed variant, no benchmark scores, and no quotas in the announcement.
• Moonshot's own figures put it at 65.7 percent on Terminal-Bench 4.0, 0.7 points behind Claude Opus 5.5 — vendor-reported, not AWS-measured.
• Try it in the Bedrock playground and measure cost and speed yourself before moving production work.

Conclusion

Kimi K3.1 Pro on Bedrock is a useful availability step, not a finished product sheet. AWS teams gain a managed route to a competitive open-weight model with a very long context window. But a launch without a price, a speed number, or a benchmark asks users to take the value on trust. Measure it yourself, watch for the Bedrock rate, and judge the model on your own work.

Sources

• AWS announcement, "Kimi K3.1 Pro (Highspeed) now available on Amazon Bedrock", 6 Nov 2026 — full reproduction at https://gixtools.net/2026/11/kimi-k3-1-pro-highspeed-now-available-on-amazon-bedrock/ (sourced to Amazon AWS)
• LetsDataScience read of the AWS announcement — https://letsdatascience.com/news/amazon-bedrock-adds-moonshot-ais-kimi-k31-pro-highspeed-mode-51717a19
• Moonshot AI, Kimi K3.1 Pro launch announcement — https://www.kimi.com/en/blog/kimi-k3-1-pro (benchmark figures)
• Amazon Web Services, "Introducing Kimi K3 on Amazon Bedrock", 18 September 2026 — https://aws.amazon.com/blogs/machine-learning/introducing-kimi-k3-on-amazon-bedrock/