Kimi prices down up to 10x, K2.7 Code HS is back
Kimi K3, K2.7 Code and K2.6 are cheaper, and K2.6 and K2.7 Code now have a cached-input rate. Kimi K2.7 Code HS is back in the catalogue.
New prices
Per 1M tokens, input / output:
Kimi K2.6 - $0.074 / $0.31, down from $0.724 / $3.354. Roughly 10x cheaper on both sides.
Kimi K2.7 Code - $0.094 / $0.42, down from $0.95 / $2.74. Input is 10x cheaper, output 6.5x.
Kimi K3 - $0.59 / $1.77, down from $0.88 / $8.74. Input is a third cheaper, output almost 5x.
Current prices for every model are on the Models and pricing page.
Cached input
On Kimi K2.6 and Kimi K2.7 Code, a repeated start of the request - the system prompt, the dialogue history, the agent context - is now charged at the lower "Cached input" rate. On Kimi K3 that rate is 9x lower than before.
Cached input per 1M tokens:
Kimi K2.6 - $0.0084
Kimi K2.7 Code - $0.009975
Kimi K2.7 Code HS - $0.01995
Kimi K3 - $0.01575, down from $0.14462
How many tokens came from cache is in the API response - the usage.prompt_tokens_details.cached_tokens field - and in the per-request breakdown under Statistics.
Kimi K2.7 Code HS is back in the catalogue
The model moonshotai/kimi-k2.7-code-highspeed is available again through the API and on the models page. It costs $0.14 per 1M input tokens, $0.90 per 1M output tokens, and $0.01995 for cached input.
Kimi K2.7 Code is text only
Kimi K2.7 Code no longer accepts images. A request with an image returns 400 unsupported_capability and is not billed. For image tasks use Kimi K3, which accepts images.
Every model that accepts images carries the "Sees images" mark on the Models and pricing page. How to pass an image in a request is covered in the documentation.