Models
Model
Kimi K2 Thinking
kimi-k2-thinking
Kimi K2 Thinking (kimi-k2-thinking) on Silicon Network: 262,144 tokens of context. Capabilities: Text, Reasoning, and Reasoning can be switched off.
| API model id | kimi-k2-thinking |
|---|---|
| Base model | Kimi K2 |
| Family | Kimi |
| Parameters | 1T MoE |
| Size class | Very large (200B and up) |
| Context length | 262,144 tokens (256K) |
| Capabilities | Text, Reasoning, and Reasoning can be switched off |
| Served weights | As published |
| Weights size | 707 GB |
| Input price | Unavailable |
| Output price | Unavailable |
The published rates could not be loaded just now. The pricing page reads them live.
Call it
The model id is kimi-k2-thinking. Any OpenAI-compatible client works with the base URL set to https://api.siliconnetwork.io/v1; the quickstart covers keys and the other wire formats.
curl https://api.siliconnetwork.io/v1/chat/completions \
-H "Authorization: Bearer $SILICON_API_KEY" \
-H "Content-Type: application/json" \
-d '{"model": "kimi-k2-thinking", "messages": [{"role": "user", "content": "Hello"}]}'Whether a model is loaded on a node right now changes through the day; the pricing page shows it live. A model that is not loaded answers 404 until an operator loads it.
Compared with
About these figures
Context length, capabilities, and weights size come from the model card each id is served with. Rates are the published per-token prices, refreshed every few minutes. There are no speed figures here: the network does not publish per-model throughput yet.