Responses

GLM5 provides a Responses-compatible endpoint for Codex Desktop, Codex CLI, and other clients that use the OpenAI Responses API.

Endpoint

POST https://glm5.app/api/v1/responses

Authenticate with a GLM5 API key:

Authorization: Bearer sk-glm5-...
Content-Type: application/json

Text response

Use glm-5.3-flash as the default model ID for new integrations:

curl https://glm5.app/api/v1/responses \
  -H "Authorization: Bearer $GLM5_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "glm-5.3-flash",
    "input": "Review this function for edge cases.",
    "max_output_tokens": 800
  }'

The response follows the standard Responses shape, including id, object: "response", status, output, output_text, model, and usage.

Input items

input can be a string or an array of standard input items. Message items with user, developer, or system roles are supported. Send the complete input history on every request; GLM5 does not persist Responses state server-side.

Streaming

Set stream to true to receive Server-Sent Events. The stream includes the standard lifecycle events, including response.created, response.output_text.delta, response.output_text.done, and response.completed.

curl --no-buffer https://glm5.app/api/v1/responses \
  -H "Authorization: Bearer $GLM5_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "glm-5.3-flash",
    "input": "Explain this error in one paragraph.",
    "stream": true
  }'

Function calling

Function tools use the Responses format. GLM5 returns function_call output items; execute the function in your client, then send a function_call_output item in the next request.

{
  "model": "glm-5.3-flash",
  "input": "Check the current project status.",
  "tools": [
    {
      "type": "function",
      "name": "shell",
      "description": "Run a safe shell command.",
      "parameters": {
        "type": "object",
        "properties": { "command": { "type": "string" } },
        "required": ["command"]
      }
    }
  ]
}

The Responses endpoint shares model availability, API keys, credits, rate limits, and spend limits with /v1/chat/completions. It does not execute client functions or provide built-in web search on your behalf.

Retry Aware Requests

Use Idempotency-Key for any request a client may retry after a network drop. The key creates one durable request record for the same API key and body. A duplicate returns the original request_id plus a status_url instead of starting another model call or another credit reservation.

Idempotency-Key: codex:turn_01JQ9KPC3H5YQ2T8

Query GET /api/v1/requests/{request_id} with the same API key to inspect whether the task is processing, succeeded, failed, or was cancelled. The endpoint exposes usage and safe failure details, but does not replay model output after a stream disconnect; GLM5 deliberately does not retain prompt or completion content for that purpose. To stop an active request, streaming or non-streaming, use POST /api/v1/requests/{request_id}/cancel; it returns 202 before the final cancelled state is written.