How to use the Kimi K3 API (pricing, context window, quickstart)
Blog post from AssemblyAI
Moonshot AI’s Kimi K3 is a 2.8-trillion-parameter open-weight model with visual understanding, always-on reasoning, and a context window listed as one million tokens on AssemblyAI’s LLM Gateway; it can also be accessed directly through Moonshot, while the Gateway offers an OpenAI-compatible API using the model ID `kimi-k3`. Gateway pricing is $3 per million input tokens and $15 per million output tokens with global routing, with a 10% surcharge for US-region requests, and the model is served by AWS Bedrock by default or Fireworks AI as an opt-in provider; both are listed as offering zero data retention, and providers do not train on customer data. Benchmark results on call transcripts show strong summaries and extraction, including a summary score of 91 on hour-long recordings, but weak long-transcript clean-up, which scored 38, alongside median latency of about 140 seconds for an hour-long transcript. The results make K3 more suitable for post-call summaries, meeting notes, and extraction than live voice interactions or lengthy transcript correction, while its cost may be higher than comparable models. Compared with Moonshot’s direct service, the Gateway has a slightly smaller listed context window but provides access to multiple models and speech-to-text through one account, making the choice dependent on context needs, provider preferences, and whether model flexibility is valuable.
Use this post, company, and trend context to find content marketing opportunities, perform competitive analysis, or address product feature gaps via the Plushcap MCP server or the Plushcap API.