Providers
Kimi Vision
Kimi K2.5, K2.6, and K3 image support with an advisory native-resolution estimate.
Current Kimi vision models
Moonshot's current Kimi API exposes multimodal models including Kimi K2.6 and Kimi K3; Kimi K2.5 is also available through the OpenAI-compatible API and accepts image/video input.
Terminal
vision-squeezer image.png --model kimi --json --dry-run
Kimi uses a native-resolution MoonViT encoder and does not publish a fixed image billing grid. VisionSqueezer therefore labels its token number advisory, while crop, output format, and upload-size savings remain exact.
Aliases: kimi, kimi-vision, kimi-k2.5, kimi-k2.6, kimi-k3.
Why the estimate is advisory
Unlike Claude, OpenAI, Gemini, Qwen, and DeepSeek's documented caps, Moonshot does not publish a stable public formula mapping image dimensions to billed tokens. The current profile uses a conservative 28px effective grid after a 4096px edge fit; do not use it as an invoice estimate.