Kimi K3 API
Pay 75% less.
On input and output tokens, compared with official prices.
OpenAI-compatible · Pay as you go
kimi-k3Kimi K3
overview.
Kimi K3 is an open-weight reasoning model from Moonshot AI. Compared with Kimi K2.6, it has a larger published context window and a higher result on the Artificial Analysis index below. On AELDER, you can compare their current token rates and use either model with the same API key and prepaid balance.
Published model specifications
OPTIONAL / CONNECTIONAPI request example
Set the model ID and base URL in your OpenAI-compatible client.
https://aelder.dev/v1import os
from openai import OpenAI
client = OpenAI(
api_key=os.environ["AELDER_API_KEY"],
base_url="https://aelder.dev/v1",
)
response = client.chat.completions.create(
model="kimi-k3",
messages=[{"role": "user", "content": "Hello!"}],
)
print(response.choices[0].message.content)import OpenAI from "openai";
const client = new OpenAI({
apiKey: process.env.AELDER_API_KEY,
baseURL: "https://aelder.dev/v1",
});
const response = await client.chat.completions.create({
model: "kimi-k3",
messages: [{ role: "user", content: "Hello!" }],
});
console.log(response.choices[0].message.content);curl https://aelder.dev/v1/chat/completions \
-H "Authorization: Bearer $AELDER_API_KEY" \
-H "Content-Type: application/json" \
-d '{
"model": "kimi-k3",
"messages": [{"role": "user", "content": "Hello!"}]
}'kimi-k3Chat CompletionsKimi K3 benchmarks.
Compare models in the AELDER catalog on the same external measure. The Artificial Analysis Intelligence Index combines evaluations of model capability. Higher is better; the score is not a percentage.
Artificial Analysis Intelligence Index
VERSION 4.3.2 · INDEX POINTS · HIGHER IS BETTER| Model | Source configuration | Index |
|---|---|---|
| Claude Opus 5.5 | Claude Opus 5.5 (Max, Default Fallback) | 58 |
| Claude Sonnet 5.5 | Claude Sonnet 5.5 (Max, Default Fallback) | 56 |
| Claude Fable 5.1 | Claude Fable 5.1 (Max, Default Fallback) | 53 |
| GPT 6 Astra | GPT-6 Astra (Max) | 53 |
| GPT 6.1 Sol | GPT-6.1 Sol (Max) | 52 |
| Claude Opus 5 | Claude Opus 5 (Max) | 51 |
| Claude Fable 5 | Claude Fable 5 (Max, Opus 4.8 Fallback) | 50 |
| GPT 6 Sol | GPT-6 Sol (Max) | 48 |
| GPT 5.6 Sol | GPT-5.6 Sol (Max) | 47 |
| DeepSeek V4.1 Flash | DeepSeek V4.1 Flash 1017 (Max) | 45 |
| GLM 5.3 | GLM-5.3 (Max) | 45 |
| Qwen3.8 Max 0902 | Qwen3.8 Max (0902) | 45 |
| Grok 4.6 | Grok 4.6 (Xhigh) | 44 |
| Kimi K3Current | Kimi K3 (Max) | 44 |
| Claude Haiku 5.5 | Claude Haiku 5.5 (Max) | 43 |
| Claude Opus 4.8 | Claude Opus 4.8 (Max) | 42 |
| GLM 5.3 Flash | GLM-5.3-Flash | 42 |
| GPT 5.6 Terra | GPT-5.6 Terra (Max) | 42 |
| Claude Opus 4.7 (1M) | Claude Opus 4.7 (Max) | 41* |
| Gemini 3.8 Flash | Gemini 3.8 Flash (High) | 41 |
| Qwen3.8 2.4T-A95B | Qwen3.8 2.4T A95B | 40 |
| Qwen3.8 Max | Qwen3.8 Max | 40 |
| Claude Sonnet 5 | Claude Sonnet 5 (Max) | 38 |
| GPT 6 Luna | GPT-6 Luna (Max) | 38 |
| GPT 5.6 Luna | GPT-5.6 Luna (Max) | 37 |
| DeepSeek V4 Pro | DeepSeek V4 Pro 0813 (Max) | 36 |
| GLM 5.2 | GLM-5.2 (Max) | 34 |
| Qwen3.8 27B | Qwen3.8 27B (Xhigh) | 34 |
| GPT 5.3 Codex | GPT-5.3 Codex (Xhigh) | 33* |
| Claude Opus 4.6 | Claude Opus 4.6 (Max) | 32* |
| Claude Sonnet 4.6 | Claude Sonnet 4.6 (Max) | 30 |
| Gemini 3.1 Pro | Gemini 3.1 Pro Preview | 30 |
| Claude Opus 4.5 | Claude Opus 4.5 (Reasoning) | 29* |
| MiniMax M3 | MiniMax-M3 | 29 |
| Kimi K2.6 | Kimi K2.6 (Reasoning) | 27 |
| GLM 5.1 | GLM-5.1 (Reasoning) | 26 |
| Kimi K2.7 Code | Kimi K2.7 Code | 26 |
| Hy3 | Hy3 | 25 |
| DeepSeek V4 Flash | DeepSeek V4 Flash 0420 (Max) · April 2026 reference | 24 |
| Kimi K2.5 | Kimi K2.5 (Reasoning) | 23* |
| MiniMax M2.7 | MiniMax-M2.7 | 23 |
DeepSeek V4 Flash uses the April 0420 release reference, labelled in the chart and table. External results refer to the labelled source configurations, which may differ from AELDER’s configuration. They do not measure AELDER API speed or uptime. Source score markers (*) are preserved.
Why are 4 catalog models missing a score?
41 of 45 catalog models have a source reference on this index version. A missing result is not a zero score. These models remain in the catalog.
Kimi K3 alternatives.
Kimi K3
questions, answered.
How do Kimi K3 and Kimi K2.6 compare on price?
Is “Kimi K3 Max” a different model I need to select?
kimi-k3 as the model ID on AELDER. “Max” identifies the reasoning configuration used for the external benchmark. The label does not mean every API request runs with that same setting.Does Kimi K3 generate images?
Your next idea,
one API away.
One key. Multiple models. Pay as you go.

