Moonshot AI launches Kimi K3 on Amazon Bedrock with 1 million token context window

Moonshot AI's new open-weights model, Kimi K3, landed on Amazon Bedrock on September 18, days after its official unveiling. It brings a 1 million token context window, native vision, and a three-tier pricing structure. The release positions K3 as a direct contender for code workloads, long-document analysis, and agents that must retain context across entire repositories.

The model is described as Moonshot's most advanced open-weights offering to date. The combination of native vision and a 1 million token window allows developers to feed large codebases, document archives, and images in a single query — no chunking, no context loss, and no improvised RAG workarounds. Moonshot targets "long-running workflows" where an agent needs to remember what it did ten thousand tokens ago and continue from there.

Running on Bedrock adds the platform's enterprise management layer: access control, encryption, auditing, and support for explicit prompt caching via the bedrock-runtime endpoint. By default the model uses automatic (implicit) caching, but configuring explicit cache controls improves hit rates and reduces both latency and cost. The official recommendation is to use the Chat Completions API; the Converse and Invoke APIs do not support the new service tiers or explicit caching.

Pricing is quoted per million tokens at a Standard baseline. The Priority tier costs 1.75 times the baseline (a 75% premium) and delivers faster response times. The Flex tier costs half the baseline (a 50% discount) and is intended for flexible, non-time-sensitive workloads. The multipliers apply to the relevant base rate — global or US CRIS — depending on the request type. Only the Chat Completions and Responses APIs support the service tiers; all other APIs are locked to Standard.

K3 is available in two cross-region profiles: US Geo, which routes only among US regions to maintain US data residency, and Global, which routes worldwide when no regulatory constraints apply. There is currently no dedicated EU or APAC profile. Bedrock's lifecycle policy applies: the model is in Active status, no end-of-life date is set, and at least 45 days' notice will be given before any deprecation.

For teams already on Bedrock, K3 adds a million-token, vision-capable option without leaving the platform — an operational advantage for compliance, managed keys, and auditing. The practical limitation right now is the absence of dedicated Europe and Asia-Pacific profiles, which may force organizations with strict data-residency requirements to wait or use the Global profile at their own risk.