How do I make the DeepSeek V3.1 model "think"?

Because this model is a "Hybrid" model, you do need to pass in an additional chat_template_kwargs={"thinking": True} configuration in your Request to the "deepseek-ai/DeepSeek-V3.1" model to turn on reasoning response generations.

An example call would look like this:

from together import Together
client = Together()

stream = client.chat.completions.create(
  model="deepseek-ai/DeepSeek-V3.1",
  messages=[
      {"role": "user", "content": "What is the most expensive sandwich?"}
  ],
  chat_template_kwargs={"thinking": True},
  stream=True,
)

for chunk in stream:
delta = chunk.choices[0].delta

# Show reasoning tokens if present
if hasattr(delta, "reasoning") and delta.reasoning:
    print(delta.reasoning, end="", flush=True)

# Show content tokens if present
if hasattr(delta, "content") and delta.content:
    print(delta.content, end="", flush=True)

Please note for this model function calling only works in non-thinking ("thinking": False) mode.