Some models think out loud before answering. They work through problems step by step, then give you a final answer. This makes them stronger at math, code, and logic-heavy tasks.
Supported models: claude-opus-45, grok-41-fast, kimi-k2-thinking, gemini-3-pro-preview, qwen3-235b-a22b-thinking-2507, qwen3-4b, deepseek-ai-DeepSeek-R1
Reading the output
Reasoning models return their thinking in one of two ways.
The reasoning_content field
Models like qwen3-235b-a22b-thinking-2507 return thinking in a separate reasoning_content field, keeping content clean:
response = client.chat.completions.create(
model="qwen3-235b-a22b-thinking-2507",
messages=[{"role": "user", "content": "What is 15% of 240?"}]
)
thinking = response.choices[0].message.reasoning_content
answer = response.choices[0].message.content
qwen3-4b, deepseek-ai-DeepSeek-R1) wrap thinking in <think> tags within the content field:
<think>
The user wants 15% of 240.
15% = 0.15
0.15 × 240 = 36
</think>
15% of 240 is **36**.
strip_thinking_response to have Venice remove them server-side.
Streaming
When streaming, reasoning_content arrives in the delta before the final answer:
stream = client.chat.completions.create(
model="qwen3-235b-a22b-thinking-2507",
messages=[{"role": "user", "content": "Explain photosynthesis"}],
stream=True
)
for chunk in stream:
if chunk.choices:
delta = chunk.choices[0].delta
if delta.reasoning_content:
print(delta.reasoning_content, end="")
if delta.content:
print(delta.content, end="")
<think> tags, the thinking streams before the answer. Collect the full response, then parse.
Reasoning effort
Reasoning models spend tokens “thinking” before they answer. The reasoning_effort parameter controls how much thinking the model does.
| Value | Behavior |
|---|
low | Minimal thinking. Fast and cheap. Best for simple factual questions. |
medium | Balanced thinking. The default for most tasks. |
high | Deep thinking. Slower and uses more tokens, but produces better answers on complex problems like math proofs or debugging. |
response = client.chat.completions.create(
model="qwen3-235b-a22b-thinking-2507",
messages=[{"role": "user", "content": "Prove that there are infinitely many primes"}],
extra_body={"reasoning_effort": "high"}
)
claude-opus-45, grok-41-fast, kimi-k2-thinking, gemini-3-pro-preview, qwen3-235b-a22b-thinking-2507
Venice also accepts the OpenRouter format: "reasoning": {"effort": "high"}. Same behavior, different syntax.
Disabling reasoning
Skip reasoning entirely for faster, cheaper responses:
response = client.chat.completions.create(
model="qwen3-4b",
messages=[{"role": "user", "content": "What's the capital of France?"}],
extra_body={"venice_parameters": {"disable_thinking": True}}
)
qwen3-235b-a22b-instruct-2507 instead.
Stripping thinking from responses
For models using <think> tags, have Venice remove them server-side:
response = client.chat.completions.create(
model="qwen3-4b",
messages=[{"role": "user", "content": "What is 15% of 240?"}],
extra_body={"venice_parameters": {"strip_thinking_response": True}}
)
qwen3-4b:strip_thinking_response=true
Parameters
| Parameter | Values | Description |
|---|
reasoning_effort | low, medium, high | Controls thinking depth |
reasoning.effort | low, medium, high | OpenRouter format |
disable_thinking | boolean | Skips reasoning entirely |
strip_thinking_response | boolean | Removes <think> tags |
disable_thinking and strip_thinking_response in venice_parameters, or use them as model suffixes.
Deprecations
qwen3-235b → qwen3-235b-a22b-thinking-2507Starting December 14, 2025, qwen3-235b routes to qwen3-235b-a22b-thinking-2507.What changes:
disable_thinking gets ignored<think> tags no longer appear in content- Thinking moves to
reasoning_content instead
What stays the same:
strip_thinking_response still works
Action required: If you parse <think> tags, switch to reading reasoning_content. If you use disable_thinking=true, switch to qwen3-235b-a22b-instruct-2507 before December 14. <think> tags will eventually be deprecated across all models in favor of the reasoning_content field.
