Run DeepSeek-R1, a reasoning model developed by DeepSeek for mathematics, software engineering, planning, and difficult multi-step analysis. Use the playground to submit complex problems, explore the model’s reasoning process, and generate detailed answers. Adjust the output settings to suit your task.
Model
deepseek-reasoner
Documentation
Use this reference to understand how every field in the playground changes the request sent to deepseek-reasoner.
Required
Type: JSON array
The conversation you want the model to continue. Enter a JSON array in chronological order. Each item needs a role and content.
[{"role":"system","content":"Answer clearly and concisely."},
{"role":"user","content":"Solve 2x + 7 = 19 and explain each step."}]Optional
Type: Boolean
Default: On
When enabled, the response is delivered progressively in small chunks as DeepSeek-R1 generates it. This provides earlier feedback and is useful for long answers. When disabled, the client waits for the complete response and receives it in one result.
Streaming changes how the result is delivered, not the question or the model’s intended answer.
Optional
Type: Integer
Default: 8,192
Range: 1–65,536
Sets the upper limit for tokens generated in this response. For DeepSeek-R1, the limit covers both its reasoning tokens and final-answer tokens, so difficult problems may need a larger allowance.
A token is a small piece of text, not exactly one word. The model can finish before reaching this limit; if the limit is too low, its reasoning or final answer may be cut off.
Optional
Type: String
Default: Text
Values: Text or JSON object
Text allows a normal natural-language answer. JSON object asks the API to return a valid JSON object, which is useful when another application will parse the result.
When choosing JSON object, also tell the model what JSON fields you expect in a system or user message and explicitly mention JSON.
Optional
Type: JSON array
Default: None
Provides one or more exact text sequences that tell the model where to stop generating. The matching sequence is not included in the returned text. Leave this field empty to let the model stop naturally or at the maximum token limit.
Enter valid JSON, even for one sequence—for example ["END"]. Matching is literal and case-sensitive, so choose sequences that are unlikely to appear accidentally.