Chat Completions

Generates a response from a list of messages, with or without streaming.

Updated on Sep 16, 2026

The core endpoint of the API. Send the history as a list of messages and receive the model's next response — the call is stateless, so the context is exactly what you sent. The guided usage is in Text generation.

POSThttps://api.hinow.ai/v1/chat/completionsBearer

Generates a response from a list of messages. Supports streaming over SSE.

Parâmetros

modelstring· bodyobrigatório
Namespaced model id, such as `hinow/himax`.
messagesarray· bodyobrigatório
The conversation messages. Content may include text and, on supported models, parts such as `image_url`.
streamboolean· body
Returns tokens as they are produced, over SSE. `usage` arrives in the last event.
temperaturenumber· body
Controls sampling variation. Lower values tend to produce more stable responses.
max_tokensinteger· body
Upper limit of tokens in the response. When it truncates, `finish_reason` comes back as `"length"`.
response_formatobject· body
`{"type": "json_object"}` requests content in JSON. Validate the fields in your application.
toolsarray· body
Tools in JSON Schema for function calling.
tool_choicestring· body
`"auto"` lets the model decide whether to call a tool.
reasoningobject· body
Reasoning control: `{ effort, enabled, exclude, max_tokens }`. See the Reasoning page.
reasoning_effortstring· body
Shortcut for `reasoning.effort`: `none`, `minimal`, `low`, `medium`, `high`, `xhigh` or `max`.
include_reasoningboolean· body
`false` hides the reasoning in the response (same as `reasoning.exclude: true`).
thinkingstring· body
`"on"` or `"off"`. Earlier alias of `reasoning.enabled`; still accepted.

Respostas

200Generated response

{
  "id": "chatcmpl-R2dSu3p8...",
  "object": "chat.completion",
  "model": "hinow/hinova",
  "choices": [
    {
      "index": 0,
      "message": { "role": "assistant", "content": "..." },
      "finish_reason": "stop"
    }
  ],
  "usage": { "prompt_tokens": 132, "completion_tokens": 60, "total_tokens": 192 }
}

400Malformed body (`invalid_request_error`)

404Model does not exist or is not enabled (`model_not_found`)

429Per-minute limit exceeded (`rate_limit_exceeded`)

Reasoning in the response

When the call asks for reasoning (or the model reasons by default), the message carries the trace next to the content, in the OpenRouter format. How to turn it on, pick the level and hide it is in Reasoning.

Field	What it says
`message.reasoning`	The reasoning as text; absent with `exclude`
`message.reasoning_details[]`	`type`, `text`, `format`, `index` and, when present, `signature`
`native_finish_reason`	The model's original stop reason
`usage.completion_tokens_details.reasoning_tokens`	Output tokens spent on reasoning
`usage.prompt_tokens_details.cached_tokens`	Input tokens served from the cache

When streaming, delta.reasoning arrives before delta.content.

finish_reason

Value	What it means	What to do
`stop`	The model finished the response	Nothing — this is the normal case
`length`	The generation hit `max_tokens`	The response is truncated; raise the limit or ask for something shorter in the prompt
`tool_calls`	The model requested a tool	Read `message.tool_calls` and return the result

Streaming and token counting

With stream: true, the usage field only appears in the last event, together with finish_reason, and the stream ends with data: [DONE]. Consume the final event when you need to record the count returned by the API.

Was this page helpful?