Chat with Kimi K3 or create an Agent
Start a direct Kimi K3 conversation for long-context coding, deep research, and visual analysis, or create a private Agent with the same fixed model.
Chat with Kimi K3 or build on it
Open the shared Kimi K3 Chatbot for a direct conversation, or create a new private Agent with a fixed Kimi K3 runtime.
Chat now or create an Agent
Start a conversation with the fixed model, or use it to create a private Agent with your own Skill and instructions.
Kimi K3 model specs and availability
Check the current API status, official model ID, release freshness, and open-weight availability before implementation.
Kimi K3 context window and key specifications
Review the operational facts most likely to affect architecture, cost, and deployment.
- Parameters
- 2.8T
- Context window
- 1,048,576 tokens
- API model ID
- kimi-k3
- Input modalities
- Text + images
- Reasoning mode
- Always on
- Maximum output
- Up to 1,048,576 tokens
Vendor-reported total parameter count.
Designed for long repositories, documents and interaction histories.
Used with Moonshot's OpenAI-compatible endpoint.
Native vision is documented for the API model.
reasoning_effort currently accepts max only.
The default documented output limit is 131,072 tokens.
Kimi K3 for coding, research, and visual workflows
Prioritize Agent workflows that can genuinely use Kimi K3's long context, native vision, and tool-oriented reasoning.
Kimi K3 benchmarks and evaluation evidence
Use Moonshot's reported evidence as a starting point, then validate Kimi K3 against your own Agent tasks before selection.
| Dimension | Available evidence | What to verify |
|---|---|---|
| Long-horizon coding | Moonshot's technical report and launch materials | Run your repository, tool loop and completion criteria |
| Long context | Documented 1,048,576-token window | Measure retrieval quality, latency and total token cost |
| Vision | Native image input documented by Moonshot | Test your image formats, detail and failure cases |
Vendor-reported evidence is not an independent production benchmark.
Kimi K3 API pricing
Review Moonshot's published Kimi K3 pricing per million tokens at the verification date.
| Token type | Price per 1M tokens | Operational note |
|---|---|---|
| Input — cache hit | $0.30 | Requires eligible automatic context caching |
| Input — cache miss | $3.00 | Use this rate for conservative input-cost estimates |
| Output | $15.00 | Reasoning and long responses can raise spend |
Confirm live pricing on Moonshot's pricing page before budgeting.
Kimi K3 limitations and deployment constraints
Account for these current API and infrastructure constraints before choosing Kimi K3 for production.
Primary sources
Official documentation used for the facts on this page.
How to use Kimi K3: FAQs
Answers about trying Kimi K3 online, pricing, context, images, open weights, local deployment, and private Agent creation.
Use Kimi K3 for your next task
Start a direct Chat now, or create a private Kimi K3 Agent for long-context coding, research, or visual work.