GPT-5 Thinking Mode

Enable GPT-5’s thinking/reasoning mode via the Responses API

GPT-5 Thinking Mode

GPT-5 supports enabling thinking mode through the Responses API’s reasoning parameter, allowing the model to perform deep reasoning before answering.

POST /v1/responses

---

Basic Usage

cURL

curl https://crazyrouter.com/v1/responses \
  -H "Content-Type: application/json" \
  -H "Authorization: Bearer YOUR_API_KEY" \
  -d '{
    "model": "gpt-5",
    "input": "Analyze the time complexity of the following code and propose an optimization:\ndef two_sum(nums, target):\n    for i in range(len(nums)):\n        for j in range(i+1, len(nums)):\n            if nums[i] + nums[j] == target:\n                return [i, j]",
    "reasoning": {
      "effort": "high"
    }
  }'

---

reasoning Parameter

Field	Type	Description
effort	string	Reasoning depth: low (quick), medium (balanced), high (deep)
summary	string	Thinking summary: auto, concise, detailed

effort Level Comparison

Level	Use Cases	Token Consumption
low	Simple questions, factual queries	Low
medium	General reasoning, code analysis	Medium
high	Complex math, deep analysis	High

---

Get Thinking Summary

Set the summary parameter to get a summary of the model’s thinking process:

Python

response = client.responses.create(
    model="gpt-5",
    input="Design a high-concurrency message queue system architecture",
    reasoning={
        "effort": "high",
        "summary": "detailed"
    }
)

# Output may contain thinking summary
for item in response.output:
    if item.type == "reasoning":
        print("Thinking process:", item.summary)
    elif item.type == "message":
        for content in item.content:
            if content.type == "output_text":
                print("Answer:", content.text)

---

Streaming Thinking

Python

stream = client.responses.create(
    model="gpt-5",
    input="Explain why the P=NP problem is important",
    reasoning={"effort": "high", "summary": "concise"},
    stream=True
)

for event in stream:
    if event.type == "response.reasoning_summary_text.delta":
        print(f"[Thinking] {event.delta}", end="")
    elif event.type == "response.output_text.delta":
        print(event.delta, end="")

---

Combined with Tools

Thinking mode can be used simultaneously with Function Calling and Web Search:

Python

response = client.responses.create(
    model="gpt-5",
    input="Analyze the current global AI chip market landscape and provide investment recommendations",
    reasoning={"effort": "high"},
    tools=[
        {"type": "web_search_preview"}
    ]
)

print(response.output_text)

---

Combined with System Instructions

Python

response = client.responses.create(
    model="gpt-5",
    instructions="You are a senior software architect who specializes in analyzing system design problems.",
    input="Design a real-time chat system that supports millions of users",
    reasoning={"effort": "high"}
)

In thinking mode, the model consumes additional tokens for internal reasoning. The higher the effort, the more tokens consumed, but the answer quality is typically better.

Not all models support the reasoning parameter. Currently it is mainly supported by GPT-5 and o-series models. For unsupported models, this parameter will be ignored.