Responses API vs Chat Completions

Both live under https://api.crazyrouter.com/v1 and use the same API key.

	Chat Completions `/v1/chat/completions`	Responses `/v1/responses`
Input	`messages` array (role/content)	`input` (string or array of items), optionally with `instructions`
Output	`choices[0].message.content`	`output[]` list of items; SDKs provide an `output_text` shortcut
Thinking/reasoning	Some models return `reasoning_content`; not stable	`output[].type = "reasoning"`, controllable via `reasoning.effort` / `summary`
Multi-turn	Resend the full `messages` every time	Continue with `previous_response_id` (must stay on the same channel; see the notes in the "Chat (Responses)" folder)
Tools	`tools` + `tool_calls`	`tools`; built-in tools such as `web_search` are supported on the model side
Compatibility	All models, all third-party clients	OpenAI-family and Codex-style clients; for Claude / Gemini use their native formats

Recommendation: use Chat Completions for general integrations and third-party tools; use Responses when you need reasoning summaries, Codex CLI, or encrypted-reasoning continuation.

Last verified/modified: 2026-09-28 (the original page mistakenly contained the "Embedding Object" content; rewritten as an actual comparison)

Responses API vs Chat API

Responses API vs Chat Completions#

Responses API vs Chat Completions