MoonshotAI: Kimi K3

Kimi K3 is a MoonshotAI model that accepts text, image, and video input and returns text. It offers a context window of 1,048,576 tokens and allows up to 943,718 completion tokens, so long documents, large codebases, or extended video-derived inputs can fit in a single request with room for lengthy outputs. It supports tool calling and reasoning; structured output support is not specified, and its weights status is not listed. The pricing is lopsided: input costs $0.239 per million tokens, while output costs $11.936 per million. That makes it most economical for workloads that read heavily and write briefly, such as long-context analysis, retrieval, or multimodal review with concise answers. Generation-heavy tasks will cost considerably more. Its blended benchmark score of 73.0 rests on a single independent benchmark (AI Index 71.9), so its standing is thinly evidenced. Teams should shortlist it for input-heavy, long-context work and validate it on their own tasks before committing.