Skip to content

[Bug] JSON Parse Error with Zhipu GLM-4.7: Stream chunks are concatenated incorrectly #7692

Closed
Closed
[Bug] JSON Parse Error with Zhipu GLM-4.7: Stream chunks are concatenated incorrectly#7692
Assignees
rekram1-node
Labels
bugSomething isn't workingwindows

Description

@yangshenlong
yangshenlong
opened on Jan 10, 2026

Description

Description: I encountered a JSON Parse Error when using the Zhipu GLM-4.7 model. It seems like the stream parser fails to handle the boundary between two data chunks correctly, leading to invalid JSON.

The Error: The error message shows that the end of the previous message and the start of the next data: chunk are stuck together without proper separation.

Error Log / Evidence:

AI_JSONParseError: JSON parsing failed: Text: ... "role":"assistant"data: {"id": ... Error message: JSON Parse error: Expected '}'

As seen in the log, the string "role":"assistant" is immediately followed by data:, missing the closing brace } and likely a newline character.

Environment:

Model: Zhipu GLM-4.7 (configured via zhipu/glm-4.7)

Plugin: OMO / OpenCode

Image

Plugins

Z.ai/GLM-4.7 ; OhMyOpencode

OpenCode version

1.0.152

Steps to reproduce

  1. Configure the OMO plugin to use Zhipu GLM-4.7 as the default model for the 'Sisyphus' agent (or set it in the model configuration).
  2. Open the OMO interface and select the 'Sisyphus' mode.
  3. Input a task to trigger the agent (e.g., "Analyze this project" or "Write a simple snake game").
  4. Wait for the model to start streaming the response.
  5. The AI_JSONParseError occurs immediately when the model returns a response, interrupting the stream.

Screenshot and/or share link

No response

Operating System

window11

Terminal

WindowsTerminal in Trae

Activity

  1. yangshenlong
    added
    bugSomething isn't working
    on Jan 10, 2026
  2. github-actions
    assigned
    rekram1-node
    on Jan 10, 2026
  3. github-actions
    added
    windows
    on Jan 10, 2026

  4. github-actions commented on Jan 10, 2026

    @github-actions
    github-actionsBot
    on Jan 10, 2026 – with GitHub Actions
    Contributor

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

    Feel free to ignore if none of these address your specific case.

  5. rekram1-node commented on Jan 11, 2026

    @rekram1-node
    rekram1-node
    on Jan 11, 2026
    Contributor

    I think this is the plugin failign to send back valid stuff when proxied through it perhaps?

  6. yangshenlong commented on Jan 11, 2026

    @yangshenlong
    yangshenlong
    on Jan 11, 2026
    Author

    I think this is the plugin failign to send back valid stuff when proxied through it perhaps?我怀疑这可能是插件在通过代理时无法返回有效数据导致的？

    Thanks for the response!

    To help you narrow down the issue, I'd like to clarify my setup details. I think this might be a specific parsing issue rather than a general network failure:

    1. Custom API Key: I am using my own API Key (configured in the zhipu provider), NOT the free built-in service provided by OpenCode. So the request should be going directly from my client to Zhipu.
    2. Proxy Environment: I am indeed using a local network proxy (VPN).
    3. Control Group (Crucial): Under the exact same proxy and custom API key settings, the Minimax model works perfectly fine. The connection is stable and the response is parsed correctly.

    The JSON Parse Error only happens with GLM-4.7. This makes me suspect that the issue is specific to how the plugin handles GLM-4.7's data stream (perhaps the plugin isn't handling concatenated JSON chunks/sticky packets correctly), while Minimax's stream format is handled fine.

    (P.S. English is not my first language, so please excuse any unnatural phrasing.

  7. DarkPocket commented on Jan 12, 2026

    @DarkPocket
    DarkPocket
    on Jan 12, 2026

    尝试用最新版本1.1.13再试试 ，
    1.0.152是旧版本

  8. rekram1-node commented on Jan 12, 2026

    @rekram1-node
    rekram1-node
    on Jan 12, 2026
    Contributor

    Oh guys make sure you are on the latest version I think that's the issue here

  9. APE-147 commented on Jan 12, 2026

    @APE-147
    APE-147
    on Jan 12, 2026

    Currently, I am using version 1.1.13, and when using it with Oh My Opencode and connecting via /connect with OpenAI OAuth, the GPT series models frequently encounter the following situation.

    Image

  10. rekram1-node commented on Jan 12, 2026

    @rekram1-node
    rekram1-node
    on Jan 12, 2026
    Contributor

    Are you running through some proxy or something that isnt handling streams correctly?

  11. APE-147 commented on Jan 13, 2026

    @APE-147
    APE-147
    on Jan 13, 2026

    After the version update, I uninstalled the opencode-openai-codex-auth plugin and switched to using /connect for OAuth authentication, which should not involve a proxy in between.

  12. rekram1-node
    marked SSE streaming parse error with GLM-4.7 responses #8282 as a duplicate of this issue on Jan 13, 2026

  13. rekram1-node commented on Jan 14, 2026

    @rekram1-node
    rekram1-node
    on Jan 14, 2026
    Contributor

    which should not involve a proxy in between.

    I meant network proxies

  14. APE-147 commented on Jan 15, 2026

    @APE-147
    APE-147
    on Jan 15, 2026

    Thank you for your patient reply.
    Based on my local environment, the system proxy on my machine: not enabled. scutil --proxy is empty; and I am directly connecting via /connect Oauth, there should be no remote reverse proxy.

  15. chindris-mihai-alexandru commented on Jan 27, 2026

    @chindris-mihai-alexandru
    chindris-mihai-alexandru
    on Jan 27, 2026

    I'm also experiencing this exact issue with Devstral-2-123B-Instruct-2512 via NVIDIA NIM provider in OpenCode v1.1.36

    Error message:

    Invalid input for tool tavily_tavily_search: JSON parsing failed: Text: {"query": "current weather in Viseu de Sus, Romania, "max_results": 1{"query": "current weather in Viseu de Sus, Romania", "max_results": 1}.
Error message: JSON Parse error: Expected '}'

    Observations:

    • The JSON payload is being duplicated/concatenated (notice the query appears twice)
    • Same stream chunk handling issue as reported for GLM-4.7
    • Not a proxy issue in my case - direct API call
    • Model loops infinitely retrying the same malformed tool call

    This suggests the stream chunk concatenation issue affects multiple providers (Z.AI GLM-4.7, NVIDIA NIM Devstral-2), not just one.

  16. github-actions commented on Mar 30, 2026

    @github-actions
    github-actionsBot
    on Mar 30, 2026 – with GitHub Actions
    Contributor

    To stay organized issues are automatically closed after 90 days of no activity. If the issue is still relevant please open a new one.

  17. github-actions
    closed this as completedon Mar 30, 2026

  18. mshareefkdy commented on Jun 28, 2026

    @mshareefkdy
    mshareefkdy
    on Jun 28, 2026

    Is there really a fix for this?

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Metadata

Metadata

Assignees

Labels

bugSomething isn't workingwindows

Type

No type

Projects

No projects

    Milestone

    No milestone

    Relationships

    None yet

    Development

    No branches or pull requests

    Issue actions