AI
tutorial
Optimize Latency by Using Non-Thinking Mode for Routine Tasks
Use the 'chat' endpoint for fast, conversational tasks and summaries.
Select the non-thinking mode (deepseek-chat) for tasks like drafting, summarizing, and general conversation to save time and tokens.
DeepSeek
The Scenario
You need to quickly summarize a long article or draft a routine email and don't require the AI to 'overthink' the logic.
Before & after
The old way
Using a heavy reasoning model for simple proofreading or email drafting can result in unnecessarily long 'thinking' pauses, taking 1-2 minutes for a simple task.
With AI
Use the deepseek-chat (non-thinking mode) for direct, low-latency responses. This reduces the 'wait time' for a response to roughly 10-30 seconds.
The Prompt
Draft a professional email based on these bullet points without using deep reasoning: [INSERT_CONTENT_HERE]
Non-thinking mode is the 'standard' mode optimized for speed and conversation rather than deep logical deliberation.
Source
Change Log | DeepSeek API Docs"deepseek-chat corresponds to DeepSeek-V3.2's non-thinking mode."