Reasoning Tokens

For models that support it, the OpenRouter API can return Reasoning Tokens, also known as thinking tokens. OpenRouter normalizes the different ways of customizing the amount of reasoning tokens that the model will use, providing a unified interface across different providers.

Reasoning tokens provide a transparent look into the reasoning steps taken by a model. Reasoning tokens are considered output tokens and charged accordingly.

Reasoning tokens are included in the response by default if the model decides to output them. Reasoning tokens will appear in the reasoning field of each message, unless you decide to exclude them.

Controlling Reasoning Tokens

You can control reasoning tokens in your requests using the reasoning parameter:

1
{
2
  "model": "your-model",
3
  "messages": [],
4
  "reasoning": {
5
    // One of the following (not both):
6
    "effort": "high", // Can be "xhigh", "high", "medium", "low", "minimal" or "none" (OpenAI-style)
7
    "max_tokens": 2000, // Specific token limit (Anthropic-style)
8
9
    // Optional: Default is false. All models support this.
10
    "exclude": false, // Set to true to exclude reasoning tokens from response
11
12
    // Or enable reasoning with the default parameters:
13
    "enabled": true // Default: inferred from `effort` or `max_tokens`
14
  }
15
}

The reasoning config object consolidates settings for controlling reasoning strength across different models. See the Note for each option below to see which models are supported and how other models will behave.

Max Tokens for Reasoning

For models that support reasoning token allocation, you can control it like this:

• "max_tokens": 2000 - Directly specifies the maximum number of tokens to use for reasoning

For models that only support reasoning.effort (see below), the max_tokens value will be used to determine the effort level.

Reasoning Effort Level

• "effort": "xhigh" - Allocates the largest portion of tokens for reasoning (approximately 95% of max_tokens)
• "effort": "high" - Allocates a large portion of tokens for reasoning (approximately 80% of max_tokens)
• "effort": "medium" - Allocates a moderate portion of tokens (approximately 50% of max_tokens)
• "effort": "low" - Allocates a smaller portion of tokens (approximately 20% of max_tokens)
• "effort": "minimal" - Allocates an even smaller portion of tokens (approximately 10% of max_tokens)
• "effort": "none" - Disables reasoning entirely

For models that only support reasoning.max_tokens, the effort level will be set based on the percentages above.

Excluding Reasoning Tokens

If you want the model to use reasoning internally but not include it in the response:

• "exclude": true - The model will still use reasoning, but it won’t be returned in the response

Reasoning tokens will appear in the reasoning field of each message.

Enable Reasoning with Default Config

To enable reasoning with the default parameters:

• "enabled": true - Enables reasoning at the “medium” effort level with no exclusions.

Examples

Basic Usage with Reasoning Tokens

1
import { OpenRouter } from '@openrouter/sdk';
2
3
const openRouter = new OpenRouter({
4
  apiKey: '{{API_KEY_REF}}',
5
});
6
7
const response = await openRouter.chat.send({
8
  model: '{{MODEL}}',
9
  messages: [
10
    {
11
      role: 'user',
12
      content: "How would you build the world's tallest skyscraper?",
13
    },
14
  ],
15
  reasoning: {
16
    effort: 'high',
17
  },
18
  stream: false,
19
});
20
21
console.log('REASONING:', response.choices[0].message.reasoning);
22
console.log('CONTENT:', response.choices[0].message.content);

Using Max Tokens for Reasoning

For models that support direct token allocation (like Anthropic models), you can specify the exact number of tokens to use for reasoning:

1
from openai import OpenAI
2
3
client = OpenAI(
4
    base_url="https://openrouter.ai/api/v1",
5
    api_key="{{API_KEY_REF}}",
6
)
7
8
response = client.chat.completions.create(
9
    model="{{MODEL}}",
10
    messages=[
11
        {"role": "user", "content": "What's the most efficient algorithm for sorting a large dataset?"}
12
    ],
13
    extra_body={
14
        "reasoning": {
15
            "max_tokens": 2000
16
        }
17
    },
18
)
19
20
msg = response.choices[0].message
21
print(getattr(msg, "reasoning", None))
22
print(getattr(msg, "content", None))

Excluding Reasoning Tokens from Response

If you want the model to use reasoning internally but not include it in the response:

1
from openai import OpenAI
2
3
client = OpenAI(
4
    base_url="https://openrouter.ai/api/v1",
5
    api_key="{{API_KEY_REF}}",
6
)
7
8
response = client.chat.completions.create(
9
    model="{{MODEL}}",
10
    messages=[
11
        {"role": "user", "content": "Explain quantum computing in simple terms."}
12
    ],
13
    extra_body={
14
        "reasoning": {
15
            "effort": "high",
16
            "exclude": True
17
        }
18
    },
19
)
20
21
msg = response.choices[0].message
22
print(getattr(msg, "content", None))

Advanced Usage: Reasoning Chain-of-Thought

This example shows how to use reasoning tokens in a more complex workflow. It injects one model’s reasoning into another model to improve its response quality:

1
from openai import OpenAI
2
3
client = OpenAI(
4
    base_url="https://openrouter.ai/api/v1",
5
    api_key="{{API_KEY_REF}}",
6
)
7
8
question = "Which is bigger: 9.11 or 9.9?"
9
10
def do_req(model: str, content: str, reasoning_config: dict | None = None):
11
    payload = {
12
        "model": model,
13
        "messages": [{"role": "user", "content": content}],
14
        "stop": "</think>",
15
    }
16
    if reasoning_config:
17
        payload.update(reasoning_config)
18
    return client.chat.completions.create(**payload)
19
20
# Get reasoning from a capable model
21
content = f"{question} Please think this through, but don't output an answer"
22
reasoning_response = do_req("deepseek/deepseek-r1", content)
23
reasoning = getattr(reasoning_response.choices[0].message, "reasoning", "")
24
25
# Let's test! Here's the naive response:
26
simple_response = do_req("~openai/gpt-mini-latest", question)
27
print(getattr(simple_response.choices[0].message, "content", None))
28
29
# Here's the response with the reasoning token injected:
30
content = f"{question}. Here is some context to help you: {reasoning}"
31
smart_response = do_req("~openai/gpt-mini-latest", content)
32
print(getattr(smart_response.choices[0].message, "content", None))

Preserving Reasoning

To preserve reasoning context across multiple turns, you can pass it back to the API in one of two ways:

1. message.reasoning (string): Pass the plaintext reasoning as a string field on the assistant message
2. message.reasoning_details (array): Pass the full reasoning_details block

Use reasoning_details when working with models that return special reasoning types (such as encrypted or summarized) - this preserves the full structure needed for those models.

For models that only return raw reasoning strings, you can use the simpler reasoning field. You can also use reasoning_content as an alias - it functions identically to reasoning.

The reasoning_details functionality works identically across all supported reasoning models. You can easily switch between OpenAI reasoning models (like ~openai/gpt-latest) and Anthropic reasoning models (like ~anthropic/claude-sonnet-latest) without changing your code structure.

Preserving reasoning blocks is useful specifically for tool calling. When models like Claude invoke tools, it is pausing its construction of a response to await external information. When tool results are returned, the model will continue building that existing response. This necessitates preserving reasoning blocks during tool use, for a couple of reasons:

Reasoning continuity: The reasoning blocks capture the model’s step-by-step reasoning that led to tool requests. When you post tool results, including the original reasoning ensures the model can continue its reasoning from where it left off.

Context maintenance: While tool results appear as user messages in the API structure, they’re part of a continuous reasoning flow. Preserving reasoning blocks maintains this conceptual flow across multiple API calls.

Example: Preserving Reasoning Blocks with OpenRouter and Claude

1
from openai import OpenAI
2
3
client = OpenAI(
4
    base_url="https://openrouter.ai/api/v1",
5
    api_key="{{API_KEY_REF}}",
6
)
7
8
# Define tools once and reuse
9
tools = [{
10
    "type": "function",
11
    "function": {
12
        "name": "get_weather",
13
        "description": "Get current weather",
14
        "parameters": {
15
            "type": "object",
16
            "properties": {
17
                "location": {"type": "string"}
18
            },
19
            "required": ["location"]
20
        }
21
    }
22
}]
23
24
# First API call with tools
25
# Note: You can use '~openai/gpt-latest' instead of '~anthropic/claude-sonnet-latest' - they're completely interchangeable
26
response = client.chat.completions.create(
27
    model="{{MODEL}}",
28
    messages=[
29
        {"role": "user", "content": "What's the weather like in Boston? Then recommend what to wear."}
30
    ],
31
    tools=tools,
32
    extra_body={"reasoning": {"max_tokens": 2000}}
33
)
34
35
# Extract the assistant message with reasoning_details
36
message = response.choices[0].message
37
38
# Preserve the complete reasoning_details when passing back
39
messages = [
40
    {"role": "user", "content": "What's the weather like in Boston? Then recommend what to wear."},
41
    {
42
        "role": "assistant",
43
        "content": message.content,
44
        "tool_calls": message.tool_calls,
45
        "reasoning_details": message.reasoning_details  # Pass back unmodified
46
    },
47
    {
48
        "role": "tool",
49
        "tool_call_id": message.tool_calls[0].id,
50
        "content": '{"temperature": 45, "condition": "rainy", "humidity": 85}'
51
    }
52
]
53
54
# Second API call - Claude continues reasoning from where it left off
55
response2 = client.chat.completions.create(
56
    model="{{MODEL}}",
57
    messages=messages,  # Includes preserved thinking blocks
58
    tools=tools
59
)

For more detailed information about thinking encryption, redacted blocks, and advanced use cases, see Anthropic’s documentation on extended thinking.

For more information about OpenAI reasoning models, see OpenAI’s reasoning documentation.

Reasoning Details API Shape

When reasoning models generate responses, the reasoning information is structured in a standardized format through the reasoning_details array. This section documents the API response structure for reasoning details in both streaming and non-streaming responses.

reasoning_details Array Structure

The reasoning_details field contains an array of reasoning detail objects. Each object in the array represents a specific piece of reasoning information and follows one of three possible types. The location of this array differs between streaming and non-streaming responses.

• Non-streaming responses: reasoning_details appears in choices[].message.reasoning_details
• Streaming responses: reasoning_details appears in choices[].delta.reasoning_details for each chunk

Common Fields

All reasoning detail objects share these common fields:

• id (string | null): Unique identifier for the reasoning detail
• format (string): The format of the reasoning detail, with possible values:
  • "unknown" - Format is not specified
  • "openai-responses-v1" - OpenAI responses format version 1
  • "azure-openai-responses-v1" - Azure OpenAI responses format version 1
  • "xai-responses-v1" - xAI responses format version 1
  • "anthropic-claude-v1" - Anthropic Claude format version 1 (default)
  • "google-gemini-v1" - Google Gemini format version 1
• index (number, optional): Sequential index of the reasoning detail

Reasoning Detail Types

1. Summary Type (reasoning.summary)

Contains a high-level summary of the reasoning process:

1
{
2
  "type": "reasoning.summary",
3
  "summary": "The model analyzed the problem by first identifying key constraints, then evaluating possible solutions...",
4
  "id": "reasoning-summary-1",
5
  "format": "anthropic-claude-v1",
6
  "index": 0
7
}

2. Encrypted Type (reasoning.encrypted)

Contains encrypted reasoning data that may be redacted or protected:

1
{
2
  "type": "reasoning.encrypted",
3
  "data": "eyJlbmNyeXB0ZWQiOiJ0cnVlIiwiY29udGVudCI6IltSRURBQ1RFRF0ifQ==",
4
  "id": "reasoning-encrypted-1",
5
  "format": "anthropic-claude-v1",
6
  "index": 1
7
}

3. Text Type (reasoning.text)

Contains raw text reasoning with optional signature verification:

1
{
2
  "type": "reasoning.text",
3
  "text": "Let me think through this step by step:\n1. First, I need to understand the user's question...",
4
  "signature": "sha256:abc123def456...",
5
  "id": "reasoning-text-1",
6
  "format": "anthropic-claude-v1",
7
  "index": 2
8
}

Response Examples

Non-Streaming Response

In non-streaming responses, reasoning_details appears in the message:

1
{
2
  "choices": [
3
    {
4
      "message": {
5
        "role": "assistant",
6
        "content": "Based on my analysis, I recommend the following approach...",
7
        "reasoning_details": [
8
          {
9
            "type": "reasoning.summary",
10
            "summary": "Analyzed the problem by breaking it into components",
11
            "id": "reasoning-summary-1",
12
            "format": "anthropic-claude-v1",
13
            "index": 0
14
          },
15
          {
16
            "type": "reasoning.text",
17
            "text": "Let me work through this systematically:\n1. First consideration...\n2. Second consideration...",
18
            "signature": null,
19
            "id": "reasoning-text-1",
20
            "format": "anthropic-claude-v1",
21
            "index": 1
22
          }
23
        ]
24
      }
25
    }
26
  ]
27
}

Streaming Response

In streaming responses, reasoning_details appears in delta chunks as the reasoning is generated:

1
{
2
  "choices": [
3
    {
4
      "delta": {
5
        "reasoning_details": [
6
          {
7
            "type": "reasoning.text",
8
            "text": "Let me think about this step by step...",
9
            "signature": null,
10
            "id": "reasoning-text-1",
11
            "format": "anthropic-claude-v1",
12
            "index": 0
13
          }
14
        ]
15
      }
16
    }
17
  ]
18
}

Streaming Behavior Notes:

• Each reasoning detail chunk is sent as it becomes available
• The reasoning_details array in each chunk may contain one or more reasoning objects
• For encrypted reasoning, the content may appear as [REDACTED] in streaming responses
• The complete reasoning sequence is built by concatenating all chunks in order

Legacy Parameters

For backward compatibility, OpenRouter still supports the following legacy parameters:

• include_reasoning: true - Equivalent to reasoning: {}
• include_reasoning: false - Equivalent to reasoning: { exclude: true }

However, we recommend using the new unified reasoning parameter for better control and future compatibility.

Provider-Specific Reasoning Implementation

Anthropic Models with Reasoning Tokens

The latest Claude models, such as ~anthropic/claude-sonnet-latest, support working with and returning reasoning tokens.

You can enable reasoning on Anthropic models only using the unified reasoning parameter with either effort or max_tokens.

Note: The :thinking variant is no longer supported for Anthropic models. Use the reasoning parameter instead.

Reasoning Max Tokens for Anthropic Models

When using Anthropic models with reasoning:

• When using the reasoning.max_tokens parameter, that value is used directly with a minimum of 1024 tokens.
• When using the reasoning.effort parameter, the budget_tokens are calculated based on the max_tokens value.

The reasoning token allocation is capped at 128,000 tokens maximum and 1024 tokens minimum. The formula for calculating the budget_tokens is: budget_tokens = max(min(max_tokens * {effort_ratio}, 128000), 1024)

effort_ratio is 0.95 for xhigh effort, 0.8 for high effort, 0.5 for medium effort, 0.2 for low effort, and 0.1 for minimal effort.

Important: max_tokens must be strictly higher than the reasoning budget to ensure there are tokens available for the final response after thinking.

Example: Streaming with Anthropic Reasoning Tokens

1
from openai import OpenAI
2
3
client = OpenAI(
4
    base_url="https://openrouter.ai/api/v1",
5
    api_key="{{API_KEY_REF}}",
6
)
7
8
def chat_completion_with_reasoning(messages):
9
    response = client.chat.completions.create(
10
        model="{{MODEL}}",
11
        messages=messages,
12
        max_tokens=10000,
13
        extra_body={
14
            "reasoning": {
15
                "max_tokens": 8000
16
            }
17
        },
18
        stream=True
19
    )
20
    return response
21
22
for chunk in chat_completion_with_reasoning([
23
    {"role": "user", "content": "What's bigger, 9.9 or 9.11?"}
24
]):
25
    if hasattr(chunk.choices[0].delta, 'reasoning_details') and chunk.choices[0].delta.reasoning_details:
26
        print(f"REASONING_DETAILS: {chunk.choices[0].delta.reasoning_details}")
27
    elif getattr(chunk.choices[0].delta, 'content', None):
28
        print(f"CONTENT: {chunk.choices[0].delta.content}")

Google Gemini 3 Models with Thinking Levels

Gemini 3 models (such as google/gemini-3.1-pro-preview and google/gemini-3-flash-preview) use Google’s thinkingLevel API instead of the older thinkingBudget API used by Gemini 2.5 models.

OpenRouter maps the reasoning.effort parameter directly to Google’s thinkingLevel values:

If a model doesn’t support a specific effort level (for example, if a model only supports low and high), OpenRouter will map your requested effort to the nearest supported level.

Using max_tokens with Gemini 3

If you specify reasoning.max_tokens explicitly, OpenRouter will pass it through as thinkingBudget to Google’s API. However, for Gemini 3 models, Google internally maps this budget value to a thinkingLevel, so you will not get precise token control. The actual token consumption is still determined by Google’s thinkingLevel implementation, not by the specific budget value you provide.

Example: Using Thinking Levels with Gemini 3

1
from openai import OpenAI
2
3
client = OpenAI(
4
    base_url="https://openrouter.ai/api/v1",
5
    api_key="{{API_KEY_REF}}",
6
)
7
8
response = client.chat.completions.create(
9
    model="{{MODEL}}",
10
    messages=[
11
        {"role": "user", "content": "Explain the implications of quantum entanglement."}
12
    ],
13
    extra_body={
14
        "reasoning": {
15
            "effort": "low"  # Maps to thinkingLevel: "low"
16
        }
17
    },
18
)
19
20
msg = response.choices[0].message
21
print(getattr(msg, "reasoning", None))
22
print(getattr(msg, "content", None))