Reasoning - Routeway Docs

Reasoning models spend additional compute “thinking through” a problem before producing a response. This extended deliberation makes them significantly better at complex tasks like multi-step math, code debugging, strategic planning, and hard classification problems where a direct answer is often wrong.

Reasoning models supported through Routeway include GPT 5.4, Claude Opus 4.8, DeepSeek V4 and much more. Check the Models page for the current list.

Enabling Reasoning

By default, Routeway disables reasoning for models. There are two ways to enable it:

Option 1: Model Suffix

Append :thinking to the end of the model ID to enable reasoning with the default effort level.

response = client.chat.completions.create(
    model="claude-opus-4-5:thinking",
    messages=[{"role": "user", "content": "Solve this step by step..."}]
)

You can also specify the effort level directly in the suffix using :high, :medium, or :low:

# High effort reasoning
response = client.chat.completions.create(
    model="o4-mini:thinking:high",
    messages=[{"role": "user", "content": "Prove that there are infinitely many primes."}]
)

# Low effort reasoning
response = client.chat.completions.create(
    model="o4-mini:thinking:low",
    messages=[{"role": "user", "content": "What is 2 + 2?"}]
)

Option 2: `reasoning_effort` Parameter

Pass the reasoning_effort parameter in the request body to enable reasoning and control the effort level:

response = client.chat.completions.create(
    model="o4-mini",
    messages=[{"role": "user", "content": "Explain the Riemann hypothesis."}],
    reasoning_effort="high"
)

Both approaches achieve the same result. The model suffix is convenient for quick testing or when there is no way to change reasoning level via ui, while reasoning_effort gives you programmatic control without changing the model ID.

When to Use Reasoning Models

Use reasoning for

Multi-step math or logic puzzles
Debugging complex code
Long-horizon planning and strategy
Hard classification with many edge cases
Tasks where chain-of-thought improves accuracy

Skip reasoning for

Simple Q&A or retrieval
Short creative tasks
High-throughput, latency-sensitive pipelines
Tasks where standard models already perform well

Controlling Reasoning Effort

You can control reasoning effort using the reasoning_effort parameter.

Value	Behavior	Best for
`"low"`	Minimal reasoning, fastest	Simple tasks, cost-sensitive
`"medium"`	Balanced (default)	Most production use cases
`"high"`	Deep reasoning, slowest	Hard math, complex code, critical decisions
`"none"`	No reasoning, fastest	Simple Q&A or retrieval

Python
Node.js
cURL

import os
from openai import OpenAI

client = OpenAI(
    base_url="https://api.routeway.ai/v1",
    api_key=os.getenv("ROUTEWAY_API_KEY")
)

response = client.chat.completions.create(
    model="o4-mini",
    messages=[
        {
            "role": "user",
            "content": "A farmer has 17 sheep. All but 9 run away. How many sheep does the farmer have left? Show your reasoning."
        }
    ],
    reasoning_effort="medium",
)

print(response.choices[0].message.content)

import OpenAI from "openai";

const client = new OpenAI({
  baseURL: "https://api.routeway.ai/v1",
  apiKey: process.env.ROUTEWAY_API_KEY,
});

async function main() {
  const response = await client.chat.completions.create({
    model: "o4-mini",
    messages: [
      {
        role: "user",
        content:
          "A farmer has 17 sheep. All but 9 run away. How many sheep does the farmer have left? Show your reasoning.",
      },
    ],
    reasoning_effort: "medium",
  });

  console.log(response.choices[0].message.content);
}

main().catch(console.error);

curl https://api.routeway.ai/v1/chat/completions \
  -H "Authorization: Bearer $ROUTEWAY_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "o4-mini",
    "messages": [
      {
        "role": "user",
        "content": "A farmer has 17 sheep. All but 9 run away. How many does he have left?"
      }
    ],
    "reasoning_effort": "medium"
  }'

​Enabling Reasoning

​Option 1: Model Suffix

​Option 2: reasoning_effort Parameter

​When to Use Reasoning Models

Use reasoning for

Skip reasoning for

​Controlling Reasoning Effort

Enabling Reasoning

Option 1: Model Suffix

Option 2: `reasoning_effort` Parameter

When to Use Reasoning Models

Controlling Reasoning Effort