Use DeepSeek-V3.2 Reasoner for Complex Agent Tasks
Enable 'thinking' mode in your API calls to handle complex agentic workflows.
Switch to the 'deepseek-reasoner' model to leverage DeepSeek-V3.2's new ability to reason during tool-use, significantly improving the reliability of autonomous agents.
The Scenario
You are building a coding assistant or a workflow agent that needs to interact with external APIs to solve multi-step problems. You want the model to 'think' through the logic before it actually executes a command.
Before & after
Previously, developers had to manually orchestrate complex agentic loops or chain multiple prompts to ensure an AI reasoned before calling a tool, often taking 30–60 minutes of debugging and prompt engineering.
By switching your API model to deepseek-reasoner, the AI natively integrates a 'thinking' phase into its tool execution, handling the multi-step logic in about 1–2 minutes.
The Prompt
Using the DeepSeek API, set your model parameter to: 'deepseek-reasoner' and provide a tool definition to solve: [DESCRIBE_COMPLEX_AGENT_TASK]
DeepSeek-V3.2 is the first model to integrate 'thinking' directly into tool-use. This means the model can reason through complex instructions before executing external functions, leading to much higher success rates in agentic workflows.
Source
Change Log | DeepSeek API Docs"DeepSeek-V3.2 is our first model to integrate thinking directly into tool-use, and also supports tool-use in both thinking and non-thinking modes."