MoonshotAI: Kimi K3 (batch)

MoonshotAI's Kimi K3 (batch) accepts text, image, and video input and returns text, with a context window of 1,048,576 tokens and a maximum completion length of 16,384 tokens. It supports tool calling and reasoning, which suits agent-style workflows and multi-step tasks. Structured output support is not listed, so teams that depend on strict schema adherence should test it before committing. The large context window allows long documents or extended video and image sets in a single request, though responses are capped well below the input limit. Pricing is $2.28 per million input tokens and $11.40 per million output tokens, which makes output the main cost driver for generation-heavy workloads. Its blended benchmark score of 73.9 rests on a single independent benchmark (AI Index score 71.9), so its standing is thinly evidenced. Shortlist it if you need very long multimodal context with tools and reasoning, but validate it on your own tasks against alternatives with broader benchmark coverage.