Loading model details
Open weights, run inside a TEE. Inference happens sealed, so the prompt stays inside the enclave.
Released
Jul 2026Modalities
Text · Tool calling · Image inputBest reported price on SayGM
-5.0%In
Cached
Out
per Mtok*
Create an API key and set it as SAYGM_API_KEY in your terminal.
curl "https://api.saygm.com/v1/chat/completions" \
-H "Authorization: Bearer $SAYGM_API_KEY" \
-H "Content-Type: application/json" \
-d '{
"model": "kimi-k3-tee",
"messages": [
{
"role": "user",
"content": "Hello!"
}
]
}'SayGM's best reported rate for kimi-k3-tee next to the published list price — 5.0% off.
Compare every Kimi model across providers in our Kimi pricing breakdown.
| Dimension | SayGM price | List price |
|---|---|---|
| Input | $2.85per Mtok | $3.00 |
| Output | $14.25per Mtok | $15.00 |
| Cache read | $0.285per Mtok | $0.30 |
What buyers have actually paid on this model recently, across every provider and cache hits.
Input
$2.85
$3.00Cached input
$0.29
$0.30Output
$14.25
$15.00Requests are load balanced across every provider serving this model, each offering its own rate, so what you pay is a blend rather than a single number. These are the blended rates actually paid — priced at what a provider offered, capped at list.
Measured over 52 windows: 268 requests and 14,583,568 input-side tokens. Most recent window closed 2026-10-10 16:28 UTC.
Every provider currently serving this model, ranked by discount.
| Provider | Discount | In | Out |
|---|---|---|---|
| Provider 1 | 5.0% off |
Explore the Moonshot AI lab and compare related models.
The best reported rate for kimi-k3-tee on SayGM is $2.85 per million input tokens and $14.25 per million output tokens.
SayGM's best reported rate for kimi-k3-tee is 5.0% off list price.
Yes. kimi-k3-tee is served on SayGM's OpenAI-compatible chat/completions surface, so an unmodified OpenAI SDK pointed at SayGM's base URL works with it.
| $2.85 |
| $14.25 |
Measured on the requests SayGM routed to this model.
Fixed-seed suites SayGM runs against this model.
What is private AI inference? →