Repository navigation
NVIDIA NIM Nemotron 3 Ultra 550B hangs indefinitely in Build/Thinking while official NVIDIA SDK works #34026
Closed as not planned
Description
Description
Summary
When using OpenCode v1.17.11 with the NVIDIA NIM provider and the model nvidia/nemotron-3-ultra-550b-a55b, the request hangs indefinitely in the Build/Thinking phase. No tokens are streamed and no error is displayed.
Environment
- OpenCode: 1.17.11
- Installation: Standalone binary (
~/.opencode/bin/opencode) - Provider: NVIDIA NIM
- Endpoint:
https://integrate.api.nvidia.com/v1 - Model:
nvidia/nemotron-3-ultra-550b-a55b - OS: Ubuntu on WSL
Configuration
{
"$schema": "https://opencode.ai/config.json",
"provider": {
"nvidia": {
"baseURL": "https://integrate.api.nvidia.com/v1",
"options": {
"extraBody": {
"chat_template_kwargs": {
"enable_thinking": true
},
"reasoning_budget": 512
},
"timeout": 600000
}
}
}
}Steps To Reproduce
- Configure the NVIDIA NIM provider.
- Select
nvidia/nemotron-3-ultra-550b-a55b. - Send a simple prompt such as
Reply with exactly: Hello!.
Expected Behavior
The model should stream reasoning and then the final response.
Actual Behavior
OpenCode remains in the Build/Thinking phase indefinitely with no streamed output and no error.
Additional Verification
Using the same API key, same endpoint, and same model, NVIDIA's official Python example (OpenAI SDK) streams and completes successfully.
Other NVIDIA NIM models also work correctly in OpenCode. The issue appears to be specific to nvidia/nemotron-3-ultra-550b-a55b.
Plugins
No response
OpenCode version
1.17.11
Steps to reproduce
- Configure the NVIDIA NIM provider.
- Select
nvidia/nemotron-3-ultra-550b-a55b. - Send a simple prompt such as
Reply with exactly: Hello!.
Screenshot and/or share link
No response
Operating System
No response
Terminal
No response
Metadata
Metadata
Assignees
Labels
No labels
Activity
github-actions commentedon Jun 26, 2026
This issue might be a duplicate of existing issues. Please check:
chat_template_kwargs/enable_thinkingrequirement for NIM reasoning modelsIf your issue is distinct (e.g., the
extraBodyworkaround doesn't resolve it for the Nemotron model specifically), please add a comment clarifying that.mahad-writes commentedon Jun 26, 2026
I'm seeing what appears to be a different issue from #24264.
Environment
nvidia/llama-3.3-nemotron-super-49b-v1(and other Nemotron models)Behavior
/connectsucceeds./v1/modelsloads correctly and models are selectable.The same NVIDIA API key and model work correctly using NVIDIA's official Python SDK against the same endpoint.
I also tested after the
extraBody-related fix discussed in #24264. The behavior is unchanged, so this does not appear to be the samechat_template_kwargsproblem affecting DeepSeek V4.This suggests there may be a separate issue affecting NVIDIA NIM/OpenAI-compatible providers, particularly Nemotron models, where the request never progresses into generation despite successful provider initialization.
mahad-writes commentedon Jun 27, 2026
@kitlangton Hello, mate. Any updates?
dyan1619 commentedon Jul 20, 2026
Hit a related but different failure mode with the same root cause (passing chat_template_kwargs/reasoning_budget for nvidia/nemotron-3-ultra-550b-a55b via NVIDIA NIM).
Instead of hanging silently, my requests get rejected outright:
Unsupported parameter(s)
Tried both:
json
"options": { "extraBody": { "chat_template_kwargs": { "enable_thinking": true }, "reasoning_budget": 512 } }
and
json
"options": { "providerOptions": { "chat_template_kwargs": { "enable_thinking": true }, "reasoning_budget": 512 } }
Both get rejected the same way — nothing in the OpenCode config schema docs indicates a hard limit on these keys, so this looks like it's coming from validation inside @ai-sdk/openai-compatible rather than OpenCode itself.
OpenCode: 1.18.3
Provider: NVIDIA NIM (https://integrate.api.nvidia.com/v1)
Model: nvidia/nemotron-3-ultra-550b-a55b
Worked around it by dropping both params entirely (model runs fine without them, just without thinking-mode control). Might be worth widening this issue's title/scope since it seems like any vendor-specific extra body param for this provider+model combo is broken, not just the hang case — possibly same underlying cause as #27462 (general request for vendor param passthrough) and #23622 (options silently dropped for openai-compatible providers).
onelpawarai commentedon Jul 20, 2026
https://github.com/onelpawarai/ZYRAXON-AI.git
fix: NVIDIA NIM model compatibility issues
fix: NVIDIA NIM model compatibility issues
github-actions commentedon Sep 20, 2026
To stay organized issues are automatically closed after 60 days of no activity. If the issue is still relevant please open a new one.