Responses

Uses the Responses protocol format — a unified surface for reasoning, agentic loops, and tool use.

POST https://wrouter.ai/v1/responses

Responses and Chat Completions cover the same conversational ground, but Responses is better for:

• Reasoning models (GPT-5, o3) with multi-step thinking & tool loops
• Agentic flows that need to interleave tool results in a stream
• Server-side session state via previous_response_id

Key parameters

Parameter	Type	Required	Notes
model	string	✓	Use Responses-capable models such as gpt-5, o3
input	string | array		Input content — string or message array
instructions	string		System-style instructions
max_output_tokens	integer		Max output tokens (note: not max_tokens)
temperature	number
top_p	number
stream	boolean		SSE
tools	array		Built-in tools (web_search, code_interpreter, …) or custom functions
tool_choice	string | object
reasoning	object		{"effort":"low"|"medium"|"high"}
previous_response_id	string		Continue from a prior response, server holds the state
truncation	string		"auto" | "disabled"

Example

from openai import OpenAI
client = OpenAI(api_key="sk-...", base_url="https://wrouter.ai/v1")

r = client.responses.create(
    model="gpt-5",
    input="Compute 2^32 and explain the reasoning.",
    reasoning={"effort": "medium"},
)
print(r.output_text)

Choosing between Chat Completions and Responses

Scenario	Prefer
Simple chat, maximum compatibility	Chat Completions
Reasoning + tool use	Responses
Cross-vendor portability	Chat Completions
Flagship reasoning models with native Responses support	Responses

For models without native Responses support, WRouter does its best to translate — but some advanced features may degrade.