Kimi K3 is available on EmpirioLabs. It is Moonshot AI's flagship reasoning model, built for software engineering, deep reasoning, and knowledge work, with a 1M token context window, always-on thinking, native web search, and text, image, and video input. It runs on an OpenAI-compatible API, so you can switch to it by changing one field.
What Kimi K3 is good at
Kimi K3 reasons before it answers on every request, so it is strong on multi-step problems: understanding a large codebase, planning and executing long agent tasks, and working through hard analytical questions. It calls tools with standard function calling, returns strict JSON on demand, reads images and video, and can search the web natively, so a single endpoint covers coding assistants, autonomous agents, and multimodal analysis.
How to call it
Send a standard chat completion. Thinking is always on, so a considered answer needs no extra setup:
curl https://api.empiriolabs.ai/v1/chat/completions \
-H "Authorization: Bearer $EMPIRIOLABS_API_KEY" \
-H "Content-Type: application/json" \
-d '{
"model": "kimi-k3",
"messages": [
{"role": "user", "content": "Refactor this function and explain the change."}
]
}'The response returns the final answer in content and the model's thinking in reasoning_content.
Tune the reasoning effort
Thinking is always on, and you can set how hard the model reasons with reasoning_effort: low, medium, high, or max. Max is the default and is recommended for complex coding and analysis:
{
"model": "kimi-k3",
"messages": [{"role": "user", "content": "Design a rate limiter and justify the tradeoffs."}],
"reasoning_effort": "max"
}Web search
Set tool_web_search to true and Kimi K3 will search the web for current information and cite its sources in the answer. A single request can run more than one search when the model needs it:
{
"model": "kimi-k3",
"messages": [{"role": "user", "content": "What changed in the latest Python release?"}],
"tool_web_search": true
}Images and video
Kimi K3 reads images and video. Pass them as content parts on a user message alongside your text, then ask a question about what they show.
Structured output
Use response_format to constrain the output. JSON mode returns any valid JSON object, and JSON schema forces output that matches a schema you provide, which is useful for extraction and tool pipelines.
Precificação
Kimi K3 is pay as you go, billed per token with no subscription, and web search is a small per-call fee only when it runs. See the live rates on the Kimi K3 model page and the pricing page, and try it in the playground.
