Kimi K3
Kimi K3 is Moonshot's general-purpose model. It reads images and reasons step by step before it answers. It takes up to 1M tokens of context and writes up to 1.05M per response.
Output price
$9.00
USD / 1M tokens
40% below list price
Add $5, get $5 more on your first top-up.
Same Kimi K3. 40% less.
Specs.
- Model ID
- kimi-k3
- Made by
- Moonshot
- Context window
- 1M tokens
- Max output
- 1.05M tokens
- Endpoint
- /v1/chat/completions
- Billing
- Per token, from one balance
Edge gateways
Requests enter at the gateway nearest your app and go straight to the model, so routing adds very little to each call.
One key, every model
Switch models by changing a string. One prepaid balance covers all of them, billed per token.
Nothing stored
We keep token counts for billing, never your prompts or responses.
Call Kimi K3 in a minute.
Point any OpenAI-compatible client at our base URL, keep your key in an environment variable, and use the model ID kimi-k3.
import os
from openai import OpenAI
client = OpenAI(
base_url="https://api.promptsforless.com/v1",
api_key=os.environ["PFL_API_KEY"],
)
response = client.chat.completions.create(
model="kimi-k3",
messages=[{"role": "user", "content": "Hello!"}],
)
print(response.choices[0].message.content)
Cursor guide
OpenCode guide
OpenAI SDK guideMore from Moonshot.
Google
AnthropicQuestions.
How much does Kimi K3 cost?
$9.00 / 1M tokens output and $1.80 / 1M tokens input, 40% below Moonshot's list price. You pay per token from a prepaid balance, with no subscription or minimum. We match your first $5 top-up, so you start with $10 of credit.
Is it the real Kimi K3?
Yes. Your requests run on Moonshot's Kimi K3, not a distilled copy or a substitute. The discount comes from how we buy capacity, not from changing the model.
Which tools can use Kimi K3?
Anything that speaks the OpenAI API: set the base URL to our endpoint and use the model ID kimi-k3. We have setup guides for Cursor, Codex, OpenCode, the OpenAI and Vercel AI SDKs, LangChain and more.
How fast is it?
Requests go through our edge gateways, so the network hop adds very little on top of Kimi K3's own generation time. Responses stream back token by token.