Environment
- OpenCode version: 1.2.22 (also tested on 1.2.24 — same behavior) - OS: Windows 11 Pro (10.0.26200)

• Installation: npm package opencode-ai, running via opencode serve
• Provider: OpenAI only, authenticated via OAuth, Team plan
• ChatGPT and Codex work fine — the OpenAI account/authentication itself has no issues
  Problem
  POST /session/{id}/message takes approximately 5 minutes to return a response, even for trivial queries like "Say hello". The response itself is correct when it eventually arrives, but the latency makes it
  unusable.

Reproduce

SESSION=$(curl -s -X POST http://localhost:4096/session | jq -r '.id')

curl -X POST "http://localhost:4096/session/$SESSION/message" \ -H "Content-Type: application/json"
-d '{
"model": {"providerID": "openai", "modelID": "gpt-4.1-mini"},
"parts": [{"type": "text", "text": "Say hello.", "role": "user"}] }'

Result: ~5 min wait, then correct response

• The same query via ChatGPT or Codex responds in seconds - Internal sessions (build agent, title generation) appear to work at normal speed

• POST /session/{id}/prompt_async has the same slow behavior
  What was tried

• Fresh DB reset (deleted opencode.db) — same result - Different models (gpt-4.1-mini, gpt-5.2) — same result

• Versions 1.2.22 and 1.2.24 — same result
  Questions
  1. Is this expected latency for the HTTP server API? 2. Is /message intended for external HTTP clients, or only for the built-in UI?

3. What is the correct way to call LLM programmatically via the HTTP server API?