> ## Documentation Index
> Fetch the complete documentation index at: https://docs.neosantara.xyz/llms.txt
> Use this file to discover all available pages before exploring further.

# Responses API Reasoning

> Control AI thinking depth using the official OpenAI SDK and reasoning effort parameters.

On the `/v1/responses` endpoint, you can calibrate model reasoning depth using the `reasoning` parameter through the official [OpenAI](https://openai.com/?utm_source=neosantara-docs\&utm_medium=referral) SDK.

## Configuring Reasoning Effort

Use the `reasoning` parameter (or `extra_body={"reasoning": ...}` in Python) to specify the model thinking budget.

<CodeGroup>
  ```python Python (OpenAI SDK) icon="python" theme={"theme":{"light":"ayu-dark","dark":"catppuccin-latte"}}
  from openai import OpenAI
  import os

  client = OpenAI(
      base_url="https://api.neosantara.xyz/v1",
      api_key=os.environ["NEOSANTARA_API_KEY"]
  )

  # Use reasoning parameter directly (or extra_body on older SDK versions)
  response = client.responses.create(
      model="deepseek-v4.1-flash",
      input="Prove mathematically why the square root of 2 is irrational.",
      reasoning={
          "effort": "high"
      }
  )

  print("Analysis Output:\n", response.output_text)
  ```

  ```javascript TypeScript (Node.js) icon="js" theme={"theme":{"light":"ayu-dark","dark":"catppuccin-latte"}}
  import OpenAI from "openai";

  const client = new OpenAI({
    baseURL: "https://api.neosantara.xyz/v1",
    apiKey: process.env.NEOSANTARA_API_KEY,
  });

  const response = await client.responses.create({
    model: "deepseek-v4.1-flash",
    input: "Prove mathematically why the square root of 2 is irrational.",
    // @ts-expect-error - responses schema extension
    reasoning: {
      effort: "high",
    },
  });

  console.log("Analysis Output:\n", response.output_text);
  ```
</CodeGroup>

## Accessing Reasoning Trace and Token Accounting

When routing to dedicated reasoning models such as `deepseek-r1`, inspect internal reasoning token usage via `usage.completion_tokens_details`.

<CodeGroup>
  ```python Python (OpenAI SDK) icon="python" theme={"theme":{"light":"ayu-dark","dark":"catppuccin-latte"}}
  from openai import OpenAI
  import os

  client = OpenAI(
      base_url="https://api.neosantara.xyz/v1",
      api_key=os.environ["NEOSANTARA_API_KEY"]
  )

  response = client.responses.create(
      model="deepseek-r1",
      input="How many trailing zeros are in 100 factorial (100!)?",
      reasoning={"effort": "high"}
  )

  # Inspect reasoning token accounting
  print(f"Completion Tokens: {response.usage.completion_tokens}")
  if hasattr(response.usage, "completion_tokens_details"):
      details = response.usage.completion_tokens_details
      print(f"Reasoning Tokens: {getattr(details, 'reasoning_tokens', 'N/A')}")

  print("\nFinal Answer:\n", response.output_text)
  ```

  ```javascript TypeScript (Node.js) icon="js" theme={"theme":{"light":"ayu-dark","dark":"catppuccin-latte"}}
  import OpenAI from "openai";

  const client = new OpenAI({
    baseURL: "https://api.neosantara.xyz/v1",
    apiKey: process.env.NEOSANTARA_API_KEY,
  });

  const response = await client.responses.create({
    model: "deepseek-r1",
    input: "How many trailing zeroes are in 100 factorial (100!)?",
    // @ts-expect-error - responses schema extension
    reasoning: { effort: "high" },
  });

  console.log(`Completion Tokens: ${response.usage?.completion_tokens}`);
  // @ts-expect-error - reasoning_tokens property
  console.log(`Reasoning Tokens: ${response.usage?.completion_tokens_details?.reasoning_tokens}`);
  console.log("\nFinal Answer:\n", response.output_text);
  ```
</CodeGroup>

## The `effort` Levels

| Value      | Description                                                                                    |
| :--------- | :--------------------------------------------------------------------------------------------- |
| `"low"`    | Minimal reasoning token budget. Recommended for structured tasks needing fast turnaround.      |
| `"medium"` | Balanced trade-off between latency and thinking depth (default).                               |
| `"high"`   | Maximum reasoning budget for mathematical proofs, complex code analysis, and multi-step logic. |

## Next Steps

| Task                        | Guide                                                         |
| :-------------------------- | :------------------------------------------------------------ |
| Anthropic Extended Thinking | [Extended Thinking](/en/gateway/anthropic-messages/thinking)  |
| OpenAI Chat Reasoning       | [Model Reasoning](/en/gateway/chat-completions/reasoning)     |
| Background Execution        | [Background Tasks](/en/gateway/responses-api/background-jobs) |


## Related topics

- [OpenResponses API](/en/gateway/responses-api.md)
- [Model Reasoning](/en/gateway/chat-completions/reasoning.md)
- [Anthropic Extended Thinking](/en/gateway/anthropic-messages/thinking.md)