Local models via Ollama take 60-90s to respond to simple prompts like 'hi', while direct API calls take 3s #18428

Description

Environment:

• OpenCode 1.2.27
• Debian GNU/Linux 13 (trixie)
• Intel Core Ultra 7 155H, Intel Arc iGPU (Vulkan)
• Ollama 0.17.7 / llama.cpp 8455

Problem:
Any local model (tested: qwen2.5-coder:3b, glm-4.7-flash, llama3.2) takes 60-90s
to respond to a simple "hi" prompt via OpenCode.

Direct API calls to the same backend return in ~3s:
time curl http://127.0.0.1:11434/v1/chat/completions ... → real 3.304s

The same delay occurs with llama-server (llama.cpp) as backend,
ruling out Ollama as the cause.

Cloud models (Anthropic, OpenAI) respond instantly — confirming
the issue is specific to local model handling inside OpenCode.

Suspected cause:
OpenCode adds ~75s of internal overhead when processing local model responses,
possibly related to event loop blocking or empty string handling in the streaming logic.

Additional symptom:
SQLiteError: SQLITE_MISUSE on startup with multiple models defined in opencode.json

Plugins

No response

OpenCode version

1.2.27

Steps to reproduce

1. Install Ollama and pull any local model:
   $ ollama pull qwen2.5-coder:3b-instruct-q4_K_M

2. Configure opencode.json with Ollama as provider:
   {
   "$schema": "https://opencode.ai/config.json",
   "provider": {
   "ollama": {
   "name": "Ollama",
   "npm": "@ai-sdk/openai-compatible",
   "options": { "baseURL": "http://127.0.0.1:11434/v1" },
   "models": {
   "qwen2.5-coder:3b-instruct-q4_K_M": {
   "name": "qwen2.5-coder:3b-instruct-q4_K_M"
   }
   }
   }
   }
   }

3. Start OpenCode and send "hi" in a new session

4. Measure response time → ~60-90s

Expected behavior:
Response within 5-10s (model warm-up) then <5s for subsequent messages

Actual behavior:
Every message takes 60-90s regardless of prompt complexity or model size

Verification that backend is not the bottleneck:
$ time curl -s http://127.0.0.1:11434/v1/chat/completions
-H "Content-Type: application/json"
-d '{"model":"qwen2.5-coder:3b-instruct-q4_K_M",
"messages":[{"role":"user","content":"hi"}],
"max_tokens":20}'
| python3 -c "import sys,json; r=json.load(sys.stdin);
print(r['choices'][0]['message']['content'])"

real 0m3.304s ← backend responds in 3s, OpenCode takes 90s

Screenshot and/or share link

No response

Operating System

debian 13 (trixie)

Terminal

QTerminal