Kimi K3 is a Kimi model tracked in Agent Serve. It supports a 1.05M token context window. Pricing starts at $3/1M input tokens and $15/1M output tokens. Key capabilities include Tool choice, Structured outputs, Reasoning low, high, max. Best for reasoning-heavy tasks that need more deliberate model control.
Best forBest for reasoning-heavy tasks that need more deliberate model control.
TemperatureNot configurable
Reasoning effortlow, high, max
VerbosityNot supported
Thinking levelsNot supported
Structured outputsSupported
Tool choiceSupported
Computer useNot supported
Deep researchNot supported
Memory supportSupported
Max output tokens1.05M
Frequently asked questions
Kimi K3 is a Kimi model available in Agent Serve. Kimi K3 is a Kimi model tracked in Agent Serve. It supports a 1.05M token context window. Pricing starts at $3/1M input tokens and $15/1M output tokens. Key capabilities include Tool choice, Structured outputs, Reasoning low, high, max.
Kimi K3 starts at $3/1M input tokens, $0.3/1M cached input tokens, and $15/1M output tokens.
Kimi K3 supports a context window of 1.05M tokens in Agent Serve. In an agent, this determines how much conversation history, tool outputs, and retrieved documents the model can hold in a single call.
Kimi K3 supports the following capabilities in Agent Serve: Tool choice, Structured outputs, Reasoning low, high, max, Max output 1.05M.
Best for reasoning-heavy tasks that need more deliberate model control. When used in an Agent Serve workflow, it can be selected in any Agent block from the model picker.
