Model inference features
Model capabilities
| Feature | Description |
|---|---|
| Thinking models | Use reasoning for complex tasks, multi-step tool calls, and agent workflows, while preserving reasoning content correctly across conversation turns. |
| ↳ Reasoning effort | Adjust reasoning depth, latency, and token usage with the levels supported by each model. |
| Multi-turn conversations | Maintain message history and control context length with a stateless API. |
| Streaming output | Receive incremental responses over SSE to reduce time to first token and process output as it arrives. |
| Multimodal input | Send text, images, and videos to supported Kimi models. |
| Structured Output | Constrain response fields and types with JSON Schema for reliable downstream parsing. |
| JSON Mode | Require a valid JSON object without constraining its exact field structure. |
| Partial Mode | Continue a given assistant prefix, fix the opening of a response, or resume truncated output. |
| Context caching | Reuse repeated request prefixes and reduce input costs with a 5m or 1h cache TTL. |
API formats
| Feature | Description |
|---|---|
| Chat Completions API | Use the OpenAI-compatible format for conversations, structured output, and tool calling. |
| Responses API | Use the OpenAI Responses-compatible format to generate text or JSON and call function tools or server-side web search. |
| Messages API | Call Kimi with the Anthropic Messages-compatible format through tools such as the Anthropic SDK and Claude Code. |
Tool calling
| Feature | Description |
|---|---|
| Custom tool calling | Define function tools, execute model-requested calls in your application, and return the results to the model. |
| ↳ Tool choice | Use tool_choice to let the model select tools automatically, require a call, or disable tool use. |
| ↳ Dynamic tool loading | Append tool definitions on demand to reduce token usage and preserve prefix caching. |
| ↳ Kimi K3 tool calling best practices | Design tool workflows with dynamic loading, tool_choice, and reasoning effort. |
| Official tools | Configure and call official Kimi Open Platform tools through the Chat Completions API. |
| Web search | Choose a search API, control costs, and improve queries and parameters for your scenario. |
| ↳ Web Search Basic | Get structured search results such as page titles, summaries, sites, and URLs through a standalone endpoint. |
| ↳ Web Search Pro | Get relevance-ranked content chunks and restrict results by site and time range. |
| URL Fetch | Fetch the title and Markdown body of a specified web page. |
Core workflows
| Feature | Description |
|---|---|
| File-based Q&A | Upload and extract file content, then add it to a conversation for single-file or multi-file Q&A. |
| ↳ Files API | Upload files and retrieve file IDs, with endpoints to list, retrieve, download, and delete files. |
| Batch API | Upload JSONL requests and asynchronously create, monitor, and download large-scale batch results. |
| Console batch inference | Create, monitor, and download batch inference results from the console without writing code. |
| Automatic reconnection | Reconnect interrupted streaming requests and use Partial Mode to resume generation. |
| Token counting | Estimate the token count of message content before sending a request. |
Integrations
| Feature | Description |
|---|---|
| Kimi Code CLI | Connect Kimi Code CLI with an Open Platform API key, switch models, and update the CLI. |
| Codex | Connect Codex to Kimi through the Responses API and verify your configuration. |
| Claude Code | Connect Kimi API to Claude Code and verify model selection and configuration. |
| OpenClaw | Connect Kimi API in OpenClaw and build cross-platform agents. |
| OpenCode | Connect Kimi models in OpenCode and configure reasoning effort. |
| Hermes Agent | Connect Kimi in Hermes Agent for text and multimodal understanding. |
| ModelScope MCP | Sync and enable ModelScope-hosted MCP servers in Playground. |
Debugging and operations
| Feature | Description |
|---|---|
| Playground | Compare models, adjust parameters, test tool calls, and convert debugging results into API requests. |
| MoonPalace | Debug Kimi API requests and manage conversations, parameters, files, tool calls, and context caching. |
| Organization, project, and API key management | Manage organization and project hierarchies, budgets, rate limits, members, IP allowlists, and usage analytics. |
| Admin API | Automate member, project, and key management, plus billing, usage, and audit queries, in your existing enterprise systems. |
| Zero Data Retention (ZDR) | Learn about the zero data retention mode available to enterprise customers. |
Related indexes
| Index | Description |
|---|---|
| Model list | View currently available models and migration guidance for retired models. |
| API overview | Review service URLs, authentication, API compatibility, and the complete endpoint list. |
| Model parameter reference | Compare default parameters, supported ranges, and usage constraints across model families. |
| Model inference pricing | Review pricing for model tokens and context caching. |
| Top-up and rate limits | Review top-up methods, rate limits, and related rules. |
| Troubleshooting | Resolve common issues with accounts, billing, authentication, model parameters, rate limits, and connectivity. |