/ api-reference / chat-completions

Chat Completions

Create a model completion — the OpenAI-compatible chat endpoint.

POST /v1/chat/completions

The core inference endpoint. Given a list of messages, the routed model returns a completion. Wire-compatible with the OpenAI Chat Completions API — OpenAI SDKs work unchanged against https://api.basemnt.ai/v1.

Request

POST/v1/chat/completions

curl https://api.basemnt.ai/v1/chat/completions \
  -H "Authorization: Bearer $SK_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "z-ai/glm-5.2",
    "messages": [
      { "role": "system", "content": "You are a terse coding assistant." },
      { "role": "user", "content": "Write a bash function that waits for a port." }
    ],
    "stream": true,
    "max_tokens": 1024
  }'

Body parameters

Parameter	Type	Default	Description
`model`	`string`	—	Required. Model ID from the catalog, e.g. `z-ai/glm-5.2`.
`messages`	`array`	—	Required. Conversation messages, oldest first. Roles: `system`, `user`, `assistant`, `tool`.
`stream`	`boolean`	`false`	Stream partial deltas via Server-Sent Events.
`max_tokens`	`integer`	model max	Cap on generated tokens (including reasoning tokens).
`temperature`	`number`	`1`	Sampling temperature, 0–2.
`tools`	`array`	—	Tool/function definitions the model may call.
`tool_choice`	`string \| object`	`auto`	`none`, `auto`, `required`, or a specific tool.
`stop`	`string \| array`	—	Up to 4 sequences where generation stops.
`response_format`	`object`	—	e.g. `{ "type": "json_object" }` for constrained JSON output (model-dependent).

Context limits

Every model in the catalog carries contextLength and maxOutputTokens. Requests whose inputs exceed the model's context are rejected with a 400 before any spend occurs.

Response (non-streaming)

200-response.json

{
  "id": "chatcmpl-8f3a...",
  "object": "chat.completion",
  "created": 1760000000,
  "model": "z-ai/glm-5.2",
  "choices": [
    {
      "index": 0,
      "message": { "role": "assistant", "content": "wait_for_port() { ... }" },
      "finish_reason": "stop"
    }
  ],
  "usage": {
    "prompt_tokens": 52,
    "completion_tokens": 184,
    "total_tokens": 236
  }
}

Streaming

With stream: true, the response is an SSE stream of chat.completion.chunk objects. Deltas arrive as choices[0].delta.content; the stream terminates on a chunk with finish_reason set, followed by data: [DONE].

sse-chunk.json

{
  "id": "chatcmpl-8f3a...",
  "object": "chat.completion.chunk",
  "model": "z-ai/glm-5.2",
  "choices": [
    {
      "index": 0,
      "delta": { "content": "wait_" },
      "finish_reason": null
    }
  ]
}

Tool-call deltas arrive on choices[0].delta.tool_calls, assembled incrementally the same way the OpenAI SDKs do — if you use an official SDK, streaming and tool calls require no extra code.

Errors

Status	Body	Meaning
`400`	`invalid_request_error`	Malformed body, unknown model, or context overflow.
`401`	`authentication_error`	Missing or invalid key. See Authentication.
`402`	`insufficient_quota`	Request-count limit reached; retry at the time in the message.
`429`	`rate_limit_error`	Requests-per-minute cap; back off and retry.
`5xx`	`api_error`	Upstream failure; retry idempotently.

Every response carries an x-request-id header — include it when contacting support.