Models
Every model has a 1M-token context window. Prices per 1M tokens are on the models page.
Available models
| Model | Modalities | Context | Reasoning |
|---|---|---|---|
Kimi K3.1 ProBest at cyberNewkimi-k3.1-pro-uncensored | 1,048,576 | low, high, maxdefault max | |
DeepSeek V4.1 Flash 1017deepseek-v4.1-flash-1017-uncensored | 1,048,576 | low, high, maxdefault high | |
DeepSeek V4.1 Flash 1017 Fastdeepseek-v4.1-flash-1017-fast-uncensored | 1,048,576 | low, high, maxdefault high | |
GLM-5.3glm-5.3-uncensored | 1,048,576 | low, high, maxdefault high | |
GLM-5.3 Flashglm-5.3-flash-uncensored | 1,048,576 | low, high, maxdefault high | |
Hy4hy4-uncensored | 1,048,576 | low, high, maxdefault high |
Images and video
Kimi K3.1 Pro takes images and video, GLM-5.3 is text only, and every other model takes images. Send images as base64: image_url parts in Chat Completions, image blocks in Anthropic Messages, or input_image parts in Responses. Send video as a base64 video_url part, in Chat Completions only. Anthropic Messages rejects video with a 400. Responses drops it, so the model answers without seeing it and the request is still billed. New accounts get a 403 with the code video_not_available until they have built some history. Images and video count as input tokens, and a whole request can be up to 64 MiB. Check anything a model reads off an image or a clip, because it can misread them.
response = client.chat.completions.create(
model="kimi-k3.1-pro-uncensored",
messages=[{"role": "user", "content": [
{"type": "text", "text": "What happens in this clip?"},
{"type": "video_url", "video_url": {"url": f"data:video/mp4;base64,{clip}"}},
]}],
)List models
GET /models returns the active models in the OpenAI format. Retired ids return 404 model_not_found.
curl https://api.silkcompute.com/models \
-H "Authorization: Bearer $SILK_API_KEY"