kimi-k3

MoonshotAdded Jul 27, 2026
Model idmoonshotai/kimi-k3

Moonshot's most capable model: an open-weight agent model that takes images as well as text, with 2.8T parameters and a 1M-token context. Built to sustain long engineering sessions with little oversight, and for agentic knowledge work such as deep research.

Cost /1M tokens
$3.00input
$0.60cached
$15.00output
Context
1Mtokens
Live speed
136 tok/s · 0.85s to first token
Speed

Throughput users got on kimi-k3 with Keln API in the last 24 hours.

100200300400 tok/s24h ago18h ago12h ago6h agonow
Latency

Time to first token users got on kimi-k3 with Keln API in the last 24 hours.

1.0 s2.0 s3.0 s4.0 s24h ago18h ago12h ago6h agonow

Capabilities

Input
Text and images
Tool calling
Yes
JSON output
Yes
Reasoning
Yes
Web search
With opt-in
Data retention
Zero

Use the API

Create an API key, save it to KELN_API_KEY, then call the endpoint below.

import os
from openai import OpenAI

client = OpenAI(
    base_url="https://api.keln.ai/v1/",
    api_key=os.environ.get("KELN_API_KEY"),
)

resp = client.chat.completions.create(
    model="moonshotai/kimi-k3",
    messages=[{"role": "user", "content": "Hello"}],
)
print(resp.choices[0].message.content)

Use with your tools

Set up guides on docs.keln.ai ↗

Ready to build on Keln?

Run kimi-k3 in the playground, or create a key and call it from your code.