Chat Completions

Request and response reference for the OpenAI-compatible GLM 5 Chat Completions API.

POST/chat/completions

Generate assistant text or function calls from a list of chat messages.

The GLM 5 Chat Completions endpoint follows the OpenAI/OpenRouter request and response shape for text chat. Existing OpenAI or OpenRouter-style clients can usually keep the same SDK and request body, then change the base URL, API key, and model ID.

https://glm5.app/api/v1

Basic Request

curl https://glm5.app/api/v1/chat/completions \
  -H "Authorization: Bearer $GLM5_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "glm-5.2",
    "messages": [
      {"role": "system", "content": "Answer in concise Markdown."},
      {"role": "user", "content": "Explain idempotency in two bullets."}
    ],
    "max_completion_tokens": 600
  }'

Request Body

Parameter	Type	Required	Default	GLM 5 behavior
`model`	string	Yes	-	Public model ID from Models, for example `glm-5.2`.
`messages`	array	Yes	-	Conversation messages. At least one valid message is required.
`max_completion_tokens`	integer	No	Model/provider	Forwarded without an application-level maximum.
`max_tokens`	integer	No	Model/provider	Legacy alias for `max_completion_tokens`; forwarded without clamping.
`temperature`	number	No	Model default	Forwarded as sampling temperature.
`top_p`	number	No	Model default	Forwarded as nucleus sampling.
`top_k`	integer	No	Model default	Forwarded when the selected upstream model supports it.
`seed`	integer	No	-	Forwarded as a deterministic sampling hint when supported.
`stop`	string or string[]	No	-	One or more stop sequences.
`stream`	boolean	No	`false`	When `true`, returns Server-Sent Events.
`tools`	array	No	-	OpenAI-compatible function tools.
`tool_choice`	string or object	No	`auto`	`auto`, `none`, `required`, or a named function choice.

The request JSON body must be at most 4 MB. GLM 5 does not impose a token-based context window or output cap before contacting the model. The selected upstream provider applies its own current limits and returns an error if the request is too large.

OpenRouter Compatibility

OpenRouter exposes a broad router API. GLM 5 accepts the common Chat Completions shape, but only the documented subset affects model behavior.

Compatible Fields

OpenRouter/OpenAI field	GLM 5 behavior
`model`	Required. Must be a GLM 5 public model ID.
`messages`	Required. Text messages and function-call history are supported.
`stream`	Supported with OpenAI-style SSE chunks and `data: [DONE]`.
`tools`	Supported for `glm-5.2`, `glm-5`, `kimi-k3`, `kimi-k2`, `deepseek-v4-pro`, and `deepseek-v4-flash`.
`tool_choice`	Supported for normal function tools.
`max_completion_tokens`	Forwarded when supplied; omitted requests use the provider default.
`max_tokens`	Forwarded as the legacy alias without an application-level cap.
`temperature`	Forwarded.
`top_p`	Forwarded.
`top_k`	Forwarded when supported upstream.
`seed`	Forwarded when supported upstream.
`stop`	Forwarded as stop sequences.

Accepted But Not Applied

Some SDKs include extra OpenRouter or OpenAI fields. GLM 5 does not currently reject every unknown field, but these fields are not forwarded to the model and should not be relied on:

Field family	GLM 5 behavior
`models`, `route`, `provider`	No model fallback chain, provider ordering, or provider filtering.
`plugins`, `transforms`	No OpenRouter web plugin, file parser, moderation, or compression.
`response_format`	JSON mode and JSON Schema modes are not currently enforced.
`frequency_penalty`, `presence_penalty`, `repetition_penalty`, `logit_bias`	Not forwarded.
`logprobs`, `top_logprobs`	Not returned.
`reasoning`, `reasoning_effort`	Not forwarded as a public API control.
`modalities`, `audio`, `image_config`	This endpoint is text-only.
`metadata`, `user`, `session_id`, `prompt_cache_key`	Not used for routing, billing, or cache affinity.
`stream_options`, `service_tier`, `prediction`, `parallel_tool_calls`	Not currently applied.

For strict integrations, remove unsupported fields before depending on their effects. For compatibility, old callers that already send harmless extra fields can continue using the endpoint.

Messages

messages is an ordered array. GLM 5 is stateless: previous turns are only remembered when your application sends them again.

Role	Content support	Purpose
`system`	String or OpenAI text parts	Application instructions and behavior constraints.
`user`	String or OpenAI text parts	User input or task data.
`assistant`	String content, `tool_calls`, or both	Prior assistant output or function calls.
`tool`	String or OpenAI text parts plus matching `tool_call_id`	Result for a previous assistant tool call.

Only text parts are used. Image, audio, video, and file parts are ignored by this text endpoint.

[
  {
    "role": "system",
    "content": "You are a concise technical assistant."
  },
  {
    "role": "user",
    "content": [
      {
        "type": "text",
        "text": "Summarize this deployment checklist."
      }
    ]
  }
]

When trimming history, keep assistant tool_calls and their matching tool messages together. A tool message without a preceding matching tool_call_id returns 400 invalid_request_error.

Non-Streaming Response

Non-streaming calls return a chat.completion object:

{
  "id": "chatcmpl-01...",
  "object": "chat.completion",
  "created": 1785292800,
  "model": "glm-5.2",
  "choices": [
    {
      "index": 0,
      "message": {
        "role": "assistant",
        "content": "- Idempotency makes retries safe..."
      },
      "finish_reason": "stop"
    }
  ],
  "usage": {
    "prompt_tokens": 42,
    "completion_tokens": 31,
    "total_tokens": 73
  }
}

Response Fields

Field	Meaning
`id`	GLM 5-generated completion ID.
`object`	Always `chat.completion` for non-streaming responses.
`created`	Unix timestamp in seconds.
`model`	Public model ID used for the request.
`choices[].index`	Choice index. GLM 5 currently returns one choice.
`choices[].message.role`	Usually `assistant`.
`choices[].message.content`	Assistant text, or `null` when the model returns only tool calls.
`choices[].message.tool_calls`	Function calls requested by the model.
`choices[].finish_reason`	`stop` or `tool_calls`.
`usage.prompt_tokens`	Input tokens reported by the provider after completion.
`usage.completion_tokens`	Output tokens reported by the provider after completion.
`usage.total_tokens`	Prompt plus completion tokens.

GLM 5 does not currently return OpenRouter-specific fields such as openrouter_metadata, usage.cost, usage.prompt_tokens_details.cached_tokens, service_tier, or system_fingerprint.

Provider Limits

GLM 5 does not return synthetic context-budget headers or reject requests using a locally configured token window. Use the final usage.prompt_tokens value for monitoring and handle any context or output-limit error returned by the selected upstream model.

Finish Reasons

Value	Meaning
`stop`	The model finished normally.
`tool_calls`	The model requested one or more function calls.

For incremental output, see Streaming. For tool-call loops, see Function calling.

On this page