Reasoning Effort

For reasoning-capable models (e.g. the GPT-5 Codex family) the gateway lets you set the reasoning effort level in three ways:

1. Model slug suffix — openai/gpt-5.1-codex:high
2. Top-level field — { "reasoning_effort": "high" } (the standard field sent by the OpenAI SDKs and OpenAI-compatible clients)
3. Nested body field — { "reasoning": { "effort": "high" } }

If more than one is provided, precedence is reasoning_effort > reasoning.effort > suffix. On Anthropic models the provider's own output_config.effort field outranks all of them.

Supported levels

Level	Suffix alias	Description
minimal	:min	Reasoning almost disabled, fastest response
low	:low	Low effort, fast
medium	:medium, :med	Default, balanced
high	:high	High effort, deeper analysis
xhigh	:xhigh, :max	Highest (on supported models only)
max	:max	On models that advertise max as its own level (e.g. openai/gpt-6-astra)

Supported levels vary by model. Sending an unsupported level returns 400 unsupported_capability.

Unknown suffixes (:turbo, :fast, etc.) return 400 invalid_request_error.

Slug suffix example

curl https://llmtr.com/v1/responses \
  -H "Authorization: Bearer llmtr-your_key" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "openai/gpt-5.3-codex:max",
    "messages": [
      {"role": "user", "content": "Make this algorithm O(n)"}
    ]
  }'

The :max suffix resolves two ways: on a model that advertises a separate max level (openai/gpt-6-astra and the GPT-5.6 family) it resolves to real max; on a model that only advertises xhigh it stays an alias for xhigh.

Body field example

curl https://llmtr.com/v1/responses \
  -H "Authorization: Bearer llmtr-your_key" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "openai/gpt-5.1-codex",
    "reasoning": { "effort": "low", "summary": "concise" },
    "messages": [
      {"role": "user", "content": "Quick one-line answer"}
    ]
  }'

summary is optional. Use it when you want a shorter or more detailed reasoning summary.

Python (OpenAI SDK)

from openai import OpenAI

client = OpenAI(
    base_url="https://llmtr.com/v1",
    api_key="llmtr-your_key",
)

response = client.responses.create(
    model="openai/gpt-5.1-codex",
    input="Refactor: make this function pure",
    reasoning={"effort": "high"},
)

print(response.output_text)

JavaScript (OpenAI SDK)

import OpenAI from "openai";

const client = new OpenAI({
  baseURL: "https://llmtr.com/v1",
  apiKey: process.env.LLMTR_API_KEY,
});

const response = await client.responses.create({
  model: "openai/gpt-5.3-codex:max",
  input: "Explain the O(n) optimization",
});

console.log(response.output_text);

Billing impact

Higher reasoning levels usually produce more output tokens and longer execution times. That can increase both cost and latency.

For more detail see the Responses API page.