Skip to content

Bug: NVIDIA API models (NIM) hang indefinitely during generation (Build/Thinking phase) despite successful model fetching #28385

Closed as not planned

Description

@trixxexe

Description

Description: When configuring a custom NVIDIA Developer API key, OpenCode successfully connects to the /v1/models endpoint and populates the UI model list. However, upon selecting any NVIDIA model and submitting a prompt, the generation hangs indefinitely in the "Build" or "Thinking" phase. No tokens are streamed to the terminal, and no error is thrown.
Meanwhile, OpenCode's built-in free models (e.g., Qwen 2.6, DeepSeek v4) work flawlessly in the exact same environment.

OS : Ubuntu (running via Termux on rootless Android 16)

Plugins

I'm using no plugins. Only the vercel theme in opencode.

OpenCode version

1.15.5

Steps to reproduce

  1. Configure OpenCode to use a custom NVIDIA API key. By connecting it by /connect
  2. Verify that the UI successfully populates the list of available NVIDIA models.
  3. Check /models to confirm
  4. Select any NVIDIA model from the list.
  5. Send a standard prompt.
  6. Observe the terminal output hanging indefinitely without streaming any response.

Screenshot and/or share link

Image

Operating System

Ubuntu (running in termux in rootless Android 16 proot-distro)

Terminal

Termux

Activity

  1. github-actions commented on May 19, 2026

    @github-actions
    Contributor

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

    If your issue is the same, please consider adding a thumbs-up or comment to the existing issue rather than opening a new one.

  2. mahad-writes commented on Jun 26, 2026

    @mahad-writes

    Additional debugging information:

    I verified that this is not an authentication or NVIDIA API issue.

    Using the same NVIDIA API key, same endpoint (https://integrate.api.nvidia.com/v1), and same model (nvidia/nemotron-3-ultra-550b-a55b), NVIDIA's official Python example using the OpenAI SDK streams successfully and returns the expected response.

    I then configured OpenCode to use the exact same API key via /connect. The model still hangs indefinitely in the Build/Thinking phase with no streamed tokens or error.

    Other NVIDIA NIM models continue to work correctly in OpenCode.

    This appears to be specific to OpenCode's handling of nvidia/nemotron-3-ultra-550b-a55b, rather than an issue with the API key, endpoint, or NVIDIA service.

    Since this issue has been linked to #24264 and #28385, it may indicate that Nemotron 3 Ultra requires additional handling beyond the previous options.extraBody propagation fix.

  3. github-actions commented on Aug 26, 2026

    @github-actions
    Contributor

    To stay organized issues are automatically closed after 60 days of no activity. If the issue is still relevant please open a new one.

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

Metadata

Metadata

Assignees

Labels

No labels
No labels

Type

No type

Projects

No projects

    Milestone

    No milestone

    Relationships

    None yet

    Development

    No branches or pull requests

    Issue actions