Amazon Web Services' Kimi K3 model card lists an 18 September 2026 launch on Amazon Bedrock, a 1 million token context window, and explicit prompt caching on the Responses and Chat Completions APIs. Inside AI, writing on 19 September, calls the listing generally available, with native vision and 2.8 trillion parameters, and says it is the first open-weight Bedrock model with that caching. AWS still marks knowledge bases and intelligent prompt routing unsupported.
The Amazon Bedrock user-guide card for Kimi K3 lists a model launch date of 18 September 2026 and a context window of 1 million tokens. It describes the model as open-weight and points to the moonshotai/Kimi-K3 card on Hugging Face. Inside AI, dated 19 September 2026, says Moonshot's Kimi K3 is generally available on Amazon Bedrock with a 1 million token context window.[1], [2]
Listed AWS capabilities include implicit prompt caching and explicit prompt caching, the latter on the Responses and Chat Completions APIs only. Inside AI frames the listing as the first open-weight Bedrock model with that caching feature.[1], [2]
AWS marks knowledge bases and intelligent prompt routing unsupported and recommends the bedrock-runtime endpoint where possible. Open weights on Bedrock are not the same as an open-source serving stack. Inside AI says access is through the Bedrock console and existing Bedrock controls, and that the model has native vision and 2.8 trillion parameters. This card stays inside the desk news cards' figures and dated stamps; it does not claim a second field census or an independent lab rerun. The reader sees the publisher's own-page development together with the second publisher's headline or lede on the same event.[1], [2]
