Try Kimi

Model Capabilities

Here are some frequently asked questions about Kimi API model capabilities.

Does it support JSON Mode?

Yes. Kimi API provides JSON Mode, which constrains model output to valid JSON. To use it:

  • Set response_format in the request parameters to {"type": "json_object"}.
  • In the prompt, clearly ask the model to return JSON and describe the expected structure.
  • In JSON Mode, the model will ensure the output is a parseable JSON string.

Does it support model fine-tuning?

Kimi API does not currently support user-initiated model fine-tuning. If you need a customized model, please contact our sales team via platform.kimi.com/contact-sales to learn about enterprise customization options.

Does it support OCR?

Kimi API's Vision models can understand images and recognize text in images, enabling OCR-like results:

  • Supports text recognition in Chinese, English, and multiple other languages.
  • Can recognize printed text and some handwriting.
  • Each image is billed at a fixed rate of 1024 tokens.
  • Beyond plain text recognition, it can also understand structured content such as tables and charts in images.

Does it support TTS (text-to-speech) and ASR (speech recognition)?

Kimi API does not currently support TTS or ASR. If you need speech-related capabilities, we recommend using Kimi API together with a third-party speech service. Please follow platform announcements for future updates.