[NIM] [GLM-5] Malformed tool-call JSON (missing '}') via OpenAI-compatible endpoint (OpenCode)

[mihai_chindris]

Hello,

I’m seeing intermittent malformed tool-call arguments when using z-ai/glm5 via NIM (https://integrate.api.nvidia.com/v1) from an OpenAI-compatible coding client called OpenCode.

Symptom
Tool calls occasionally contain truncated/malformed JSON args, e.g. missing closing }:

Invalid input for tool tavily_tavily_search: JSON parsing failed: Text: {“query”:“Geoffrey Huntley LinkedIn Ralph Wiggum most viral reposts”.
Error message: JSON Parse error: Expected ‘}’

In the same session, multiple tools can fail similarly (tavily_tavily_search, brave-search_brave_web_search, webfetch), and the client may enter retry loops.

Environment

• Model: z-ai/glm5
• Endpoint: https://integrate.api.nvidia.com/v1
• Client: OpenCode 1.2.6
• OS: macOS Sequoia 15.7.4

Reproduction

1. Configure OpenAI-compatible client to use NIM endpoint + z-ai/glm5.
2. Run a prompt that triggers several tool/function calls in one turn.
3. Observe intermittent malformed tool JSON payloads.

Expected
Tool/function call arguments are always strict valid JSON.

Actual
Occasional malformed JSON in tool arguments causes parser errors and failed tool execution.

Related reports

• OpenCode issue: [BUG] GLM-5 via NVIDIA NIM emits malformed MCP tool JSON (Expected '}') in OpenCode v1.2.6 · Issue #13900 · anomalyco/opencode · GitHub
• GLM-5 issue: [Bug] Malformed tool-call JSON via NVIDIA NIM (GLM-5) causes parser failures in OpenCode · Issue #15 · zai-org/GLM-5 · GitHub
• vLLM issue (similar malformed tool-calls): [Bug]: GLM-5-FP8 malformed tool calls · Issue #34449 · vllm-project/vllm · GitHub

If there is a recommended parser/config/flag set for robust GLM-5 tool-calling via NIM, please share.

Thanks!

[mihai_chindris]

Update: minimal repro + exact malformed tool-call payloads here:

https://gist.github.com/chindris-mihai-alexandru/f3c114915f18618406030e6a2d8fa59b

[calmcrazy06]

Me too.I found that GLM-5 cannot correctly output JSON in json_object mode.

[mihai_chindris]

GLM-5 via OpenCode (NIM OpenAI endpoint) still returning EngineCore encountered an issue

Environment:

• macOS 26.3
• OpenCode CLI 1.2.15
• Model: z-ai/glm5 via https://integrate.api.nvidia.com/v1
• Request ID (from direct API call): 83cc2550-a234-445d-b5b4-7da6b35e4497

Steps:

1. Run opencode run --model nvidia/z-ai/glm5 "Quick smoke test".

2. OpenCode log line 115 shows llm providerID=nvidia modelID=z-ai/glm5, but the CLI never receives any tokens before timing out.

3. To isolate OpenCode from the provider:

   curl -s -D nim-glm5-headers.txt -o nim-glm5-body.json \
     -H "Authorization: Bearer $NVIDIA_API_KEY" \
     -H "Content-Type: application/json" \
     -X POST https://integrate.api.nvidia.com/v1/chat/completions \
     -d '{"model":"z-ai/glm5","messages":[{"role":"user","content":"ping"}],"max_tokens":64}'
   

4. The API immediately returns HTTP/2 500 and the body {“error”:{“message”:“EngineCore encountered an issue. See stack trace (above) for the root cause.”,“type”:“InternalServerError”…}}.

Artifacts: opencode-log-2026-03-05T224023.log, nim-glm5-headers.txt, nim-glm5-body.json.

glm5-nim-error.zip (5.2 KB)