MoonshotAI: Kimi K3

Kimi K3 is a MoonshotAI model that accepts text, image, and video input. It offers a context window of 1,048,576 tokens and allows up to 943,718 completion tokens, so very long documents and long outputs fit within a single request. It supports tool calling and reasoning. The listing does not confirm structured output support or whether the weights are open, so verify both before building around either. Pricing is lopsided: $0.73 per million input tokens is inexpensive, while $13.5 per million output tokens is steep. That makes the model most cost-effective for input-heavy work, such as analyzing large files, long transcripts, or video, where responses stay short. Its blended benchmark score of 73.9 comes from a single independent benchmark, so its standing is thinly evidenced. Shortlist it if you need multimodal input and a very large context, and run your own tests before committing to output-heavy workloads.