Repository navigation
Vultr provider model selection not respected #18352
Description
Description
I installed OpenCode desktop and configured Vultr provider, enabling GPTOSS-120 model.
After running few queries, I checked my vultr usage page and noticed that MiniMax-M2.5 model was used.
I tried that with another model and saw the same result.
As a workaround, I configure Vultr as custom provider and select the desired model.
Plugins
No response
OpenCode version
1.2.27
Steps to reproduce
- Configure Vultr provider (set API key)
- Under Models, Enable Vultr/GPTOSS-120
- Select Vultr/GPTOSS-120 from the list of models in the main window.
- Run some query
- Check Vultr portal for usage - It shows that MiniMax-M2.5 was used and not usage appears for GPTOSS-120
Screenshot and/or share link
No response
Operating System
Ubuntu 24.04
Terminal
No response
Metadata
Metadata
Assignees
Labels
No labels
Activity
github-actions commentedon Mar 20, 2026
This issue might be a duplicate of (or related to) existing issues. Please check:
scaleway/gpt-oss-120bunable to call tools — Vultr's GPTOSS-120 is Scaleway'sgpt-oss-120bmodel; this issue covers tool-calling failures with that model via the chat/completions endpoint vs. the responses API, which may be related to why the provider falls back to a different model (MiniMax-M2.5)biondizzle commentedon Apr 24, 2026
This wasn't an OpenCode bug. OpenCode sends the model name as lower case. Our model look up was case sensitive. When we don't have the model requested, we default to MiniMax. We fixed this by making the model look up case insensitive. Why do we default to minimax? It fits comfortably on a minimum of 4 cards and the most stable model (its architecture so far just works out of the box with every optimization stably i.e. turboquant, LMCache, etc)
As we shuffle other models around during roll outs, it doesnt need to take up a whole b200 node on its own so we always have a replica of it live during these windows when something like an optimization for a different model crashes after roll out was marked successful. So when the lookup fails, it just defaults to minimax to keep requests flowing.
Although we recently had to take down GPT OSS due to capacity constraints after the rush of new models during the last 2 weeks so if you're still calling that, it will default to minimax. We'll be bringing it back hopefully soon.
github-actions commentedon Jul 4, 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.