JSON parsing failed: Text: - macOS / Docker

[BastianZuehlke]

Description

I randomly get JSON parsing errors, then the processing stops.

I get the same issue even if the LLM get served from Ollama or LM-Studio.
It happens if I use OpenCode directly on macOS or using the docker image.

I am using a clean OpenCode install, no tools (other than the default ones) or MCPs.

JSON parsing failed: Text: {"id":"chatcmpl-8w4gtjb6tplr8g1xoxrjvi","object":"chat.completion.chunk","created":1777577130,"model":"qwen/qwen3.6-27b","system_fingerprint":"qwen/qwen3.6-27b","choices":[{"index":0,"delta":{"reasoning_content":"\n"},"logprodata: {"id":"chatcmpl-b7yoftfoinwhk4joyloyjt","object":"chat.completion.chunk","created":1777577321,"model":"qwen/qwen3.6-27b","system_fingerprint":"qwen/qwen3.6-27b","choices":[{"index":0,"delta":{"role":"assistant","reasoning_content":"Let"},"logprobs":null,"finish_reason":null}]}.
Error message: JSON Parse error: Expected ':' before value in object property definition

Plugins

None

OpenCode version

1.14.30 (macOS) , 1.14.25 (Docker)

Steps to reproduce

Just need to give any prompt.

Screenshot and/or share link

No response

Operating System

macOS (npm install), docker

Terminal

No response

[github-actions]

This issue might be a duplicate of existing issues. Please check:

• SSE JSON parse failure with GLM-5.1 (Z.AI) — malformed JSON in content field kills stream #23442: Same root cause — model hallucinates SSE-formatted chunks in its response content, producing malformed JSON that kills the stream. The error pattern and fix described there match your report closely.

If your issue is distinct (e.g. a different reproduction path or different root cause), please add more context to differentiate it.

[BastianZuehlke]

My understanding (based on reading the other issues) the llm is returning this corrupt JSON which stops OpenCode execution. ?
I tested qwen 3.6 27b and 30b, so these models should / can not be used in OpenCode, at least not reliably ?

[water-in-stone]

My understanding (based on reading the other issues) the llm is returning this corrupt JSON which stops OpenCode execution. ? I tested qwen 3.6 27b and 30b, so these models should / can not be used in OpenCode, at least not reliably ?

I have submmited a MR to fix this bug.