Moonshot AI: Kimi K3

Kimi K3 is Moonshot AI's flagship open-weight multimodal reasoning model. It is designed for complex coding, knowledge work, and long-horizon agentic workflows, and is especially strong at large repository understanding, tool use, debugging, and iterative work across images, logs, tests, and runtime feedback.

WaveSpeed AI exposes moonshotai/kimi-k3 through an OpenAI-compatible API, so it can be used with standard OpenAI SDKs and existing chat-completions-based application flows.

---

Why Use Kimi K3

• Flagship Kimi model for advanced reasoning and software work
• Strong long-horizon coding performance across large repositories and multi-step tasks
• Well suited for agentic workflows, tool use, and structured outputs
• Native multimodal capability for image-based understanding and visual iteration
• Long-context support for document analysis, code review, and extended multi-turn sessions

Key Features

• Context Window: 1,000,000 tokens
• Max Output: up to 131,072 tokens by default
• Vision Input: Supported
• Function Calling: Supported
• Structured Outputs: Supported
• Reasoning: Enabled by default
• Best Fit: coding, reasoning, agents, multimodal workflows, long-context tasks

Specifications

---

Architecture Notes

Kimi K3 is built with KDA (Kimi Delta Attention) and Attention Residuals to improve computational efficiency at scale. It is positioned for demanding workflows such as long-horizon programming, knowledge-intensive tasks, and multimodal reasoning over both text and images.

---

How to Use

Chat Completions

Use Chat Completions when you want a straightforward OpenAI-compatible integration path for conversational, coding, and agent workflows.

Python

from openai import OpenAI

client = OpenAI(
    api_key="YOUR_API_KEY",
    base_url="https://llm.wavespeed.ai/v1"
)

response = client.chat.completions.create(
    model="moonshotai/kimi-k3",
    messages=[
        {"role": "user", "content": "Review this bug report and identify the most likely root cause."}
    ]
)

print(response.choices[0].message.content)

cURL

curl https://llm.wavespeed.ai/v1/chat/completions \
  -H "Content-Type: application/json" \
  -H "Authorization: Bearer YOUR_API_KEY" \
  -d '{
    "model": "moonshotai/kimi-k3",
    "messages": [
      {"role": "user", "content": "Review this bug report and identify the most likely root cause."}
    ]
  }'

---

Multimodal Example

Kimi K3 supports native visual understanding, making it a strong fit for tasks such as screenshot debugging, UI review, diagram analysis, and image-grounded reasoning.

Python

from openai import OpenAI

client = OpenAI(
    api_key="YOUR_API_KEY",
    base_url="https://llm.wavespeed.ai/v1"
)

response = client.chat.completions.create(
    model="moonshotai/kimi-k3",
    messages=[
        {
            "role": "user",
            "content": [
                {"type": "text", "text": "Describe the issue shown in this screenshot."},
                {
                    "type": "image_url",
                    "image_url": {
                        "url": "data:image/png;base64,BASE64_IMAGE_DATA"
                    }
                }
            ]
        }
    ]
)

print(response.choices[0].message.content)

---

Tool Use and Structured Output

Kimi K3 is well suited for applications that combine reasoning with tools and schema-constrained outputs.

Common use cases include:

• debugging agents that inspect logs, test output, and code
• repository assistants that navigate large codebases
• multimodal workflows that combine screenshots with implementation tasks
• structured extraction pipelines that require JSON output

---

Notes

• Official upstream model name is kimi-k3
• WaveSpeed model ID is drafted here as moonshotai/kimi-k3
• Reasoning is part of the model's default behavior
• Best paired with agentic and long-context workflows where tool use and iterative refinement matter