Moonshot AI's Kimi K3 now generally available on Amazon Bedrock

AWS made Moonshot AI's Kimi K3 generally available on Amazon Bedrock, positioning it as a powerful open-weight choice for coding and knowledge work. Moonshot describes it as its most capable model to date and claims it is the first open model to reach 2.8 trillion parameters, paired with native vision, a 1-million-token context window, and explicit prompt caching for cost control on long contexts. The parameter count, if accurate, makes it one of the largest open-weight models publicly available.

The Kimi K3 launch was part of a broader Bedrock roster expansion: AWS also added OpenAI's GPT-6 Astra and Anthropic's Claude Fable 5.1, reinforcing Bedrock's multi-model, best-of-breed positioning against Azure AI Foundry and Google Vertex. The strategy is to be the neutral marketplace where enterprises can mix frontier proprietary models with the newest open-weight releases without switching platforms.

Kimi K3's arrival extends the week's clear theme of 1M-token omni-modal and large open models flooding the market — alongside Alibaba's Qwen3.8-Omni-Flash and the general race between context length and true multimodal reasoning that has developers debating which actually matters. For AWS customers, the appeal is running a 2.8T-parameter open model with managed serving, autoscaling, and prompt caching rather than self-hosting the enormous infrastructure such a model demands. The caveat is cost and latency at that scale; explicit prompt caching helps, but a 2.8T-parameter model is not cheap to run, and buyers will want to see whether its coding and vision benchmarks justify the footprint versus smaller, faster alternatives.