[bug] GLM 5 keeps screwing up the json parsing of read tool calling

[YukariMindGap]

Description

I've been using GLM 5 from NanoGPT for a while, running opencode cli inside VS Code. But lately there's an increase of tendency for the ai to get stuck in a loop due to tool calling errors, mostly on the "read" tool. I have no idea what's happening but apparently the ai keeps screwing up the JSON formatting for the tool calling. Sometimes i even get "Method Not Allowed" errors that stops the build dead in the tracks. I thought it's the Superpowers plugin i've installed but i've made a clean reinstall and the errors persist.

I've attached some error logs while testing on a test project.

Plugins

None. it's a fresh install

OpenCode version

1.2.6

Steps to reproduce

1. Go to a provider that offers GLM-5 (I use NanoGPT for this instance, but i think other providers might also be affected)
2. Switch model to GLM-5 Thinking (Haven't tested in non-thinking one and i think other chinese models might have this problem)
3. Create a test project with some files with contents inside
4. Ask the ai to try reading those files and summarize them to you.
5. Watch the reasoning logs and the occasional build-stopping error.

Screenshot and/or share link

Operating System

Windows 11

Terminal

Windows Terminal in VS Code

[github-actions]

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

• [BUG] GLM-5 via NVIDIA NIM emits malformed MCP tool JSON (Expected '}') in OpenCode v1.2.6 #13900: GLM-5 via NVIDIA NIM emits malformed MCP tool JSON (Expected '}') in OpenCode v1.2.6
• 🐛 Bug Report: MCP Tool JSON Parsing Errors on OpenCode (GLM-4.7) #8102: MCP Tool JSON Parsing Errors on OpenCode (GLM-4.7)
• [Bug] JSON Parse Error with Zhipu GLM-4.7: Stream chunks are concatenated incorrectly #7692: JSON Parse Error with Zhipu GLM-4.7: Stream chunks are concatenated incorrectly

These issues describe very similar symptoms with GLM models producing malformed JSON during tool calls. The #13900 issue is particularly similar as it also reports GLM-5 with malformed JSON errors in v1.2.6.

[YukariMindGap]

I tried downgrading from 1.2.6 to 1.2.4 but the error still persists

[rekram1-node]

@YukariMindGap can u send me the session?

opencode export > session.json

RUN this exact command in your terminal, select the session and send me the json file from the output.

[YukariMindGap]

@YukariMindGap can u send me the session?

opencode export > session.json

RUN this exact command in your terminal, select the session and send me the json file from the output.

session.json

here it is

@rekram1-node

[YukariMindGap]

I also get this problem from Qwen 3.5 and Kimi 2.5, so i think that might be an issue of the models themselves. Haven't tried on GPT 5.2, Claude 4.5 Opus/ 4.6 Sonnet, or Gemini 3 Pro.

[flyboat403]

I also get same problem

[YukariMindGap]

I also get this problem from Qwen 3.5 and Kimi 2.5, so i think that might be an issue of the models themselves. Haven't tried on GPT 5.2, Claude 4.5 Opus/ 4.6 Sonnet, or Gemini 3 Pro.

Minimax 2.5, oddly enough, doesn't have this problem. I was testing Qwen 3.5 and Kimi 2.5 with the thinking model, so that might be a problem

[YukariMindGap]

I think the malformed json has been mostly fixed by itself, but now i got this problem where tool calling keeps coming up with "missing". I've found no reason why this happened.