moonshotai

/

Kimi-K2.5

like 2.18k

Follow

Moonshot AI 6.85k

Kimi-K2.5 is a hybrid thinking model that supports both thinking mode and instant(disable thinking) mode. You can refer to https://huggingface.co/moonshotai/Kimi-K2.5#chat-completion to see how to enable/disable thinking.

Hi @Lord-Shibe ,

Based on the README and model documentation:

Kimi K2.5 supports both modes in a single model:

1. Thinking Mode (default): The model uses <think>...</think> tags to show its reasoning process before providing the final answer. This is enabled by default.

   • Temperature: 1.0 recommended
   • Access reasoning via response.choices[0].message.reasoning_content

2. Instant Mode: Direct response without explicit reasoning. Enable by passing:

   # Official API
   extra_body={'thinking': {'type': 'disabled'}}
   
   # vLLM/SGLang
   extra_body={'chat_template_kwargs': {"thinking": False}}
   

   • Temperature: 0.6 recommended

So "K2.5" and "K2.5 Thinking" refer to the same model - it's just a matter of which mode you're using. The model was trained with both capabilities built-in.

Hope this helps!