OpenCode hangs indefinitely with @ai-sdk/openai-compatible although backend completes successfully

Description

Environment

• OS: Windows
• OpenCode: 1.18.25
• Provider: 9router (@ai-sdk/openai-compatible)
• 9router: 0.5.59

---

Problem

OpenCode never finishes responding.

The UI remains in the loading state forever.

However, the backend successfully completes the request.

---

Reproduction

1. Configure OpenCode to use 9router.
2. Send a simple prompt ("Hello").
3. Wait.

Result:

• Spinner never stops.
• No error appears.
• No response is rendered.

---

Backend evidence

9router logs show:

• request received
• model fallback works
• final model succeeds
• DONE

So the backend successfully finishes the request.

---

Investigation

I inspected OpenCode source code.

Relevant findings:

• provider/provider.ts
• session/llm.ts
• session/llm/ai-sdk.ts
• session/processor.ts

Spinner is not stopped when the "finish" event arrives.

Instead it only stops after:

Stream.runDrain()
→ cleanup()
→ Runner.onIdle()
→ status = idle

Therefore if fullStream never finishes, the UI remains loading forever.

---

Possible cause

For @ai-sdk/openai-compatible providers, OpenCode automatically injects:

includeUsage = true

If the backend stream does not produce the expected usage event (or produces it in an incompatible format), fullStream may never complete.

This matches the observed behavior:

Backend:
DONE

Frontend:
Loading forever.

---

Expected

OpenCode should exit loading once the backend has completed the stream.

---

Additional information

I already confirmed:

• backend receives request
• backend completes request
• backend is reachable
• issue persists with different models

Therefore the issue appears to be in the OpenCode streaming lifecycle rather than request dispatch.

Plugins

No response

OpenCode version

No response

Steps to reproduce

No response

Screenshot and/or share link

No response

Operating System

No response

Terminal

No response