[BUG] Starting from version v3.33.2, the model ID that actually exists cannot be found in the openrouter provider, and this bug still exists in the latest version v3.34.4.

[zx2021210538]

Problem (one or two sentences)

Starting from version v3.33.2, the model ID that actually exists cannot be found in the openrouter provider, and this bug still exists in the latest version v3.34.4.
Currently, I have identified that this bug occurs in the GPT-5.1 series models, Gemini-3 model, and Grok-4.1-Fast model. It is not yet known whether other models also have the same bug.
An interesting phenomenon is that in the first few minutes after you just open Roo Code, this bug will not appear, and you can normally configure the models in OpenRouter (such as the GPT-5.1 series). However, after some time, when you open the model configuration page of Roo Code again, it will show that the GPT-5.1 series models do not exist, and the models will be automatically replaced with Claude Sonnet 4.5.

Context (who is affected and when)

In all versions subsequent to v3.33.1, this bug will occur when you configure some models from OpenRouter. The "some models" here refer to the gpt-5.1 series, gemini-3, and grok-4.1-fast.
An interesting phenomenon is that in the first few minutes after you just open Roo Code, this bug will not appear, and you can normally configure the models in OpenRouter (such as the GPT-5.1 series). However, after some time, when you open the model configuration page of Roo Code again, it will show that the GPT-5.1 series models do not exist, and the models will be automatically replaced with Claude Sonnet 4.5.

Reproduction steps

1. Open the settings interface of Roo Code;
2. Create a new configuration, select the "openrouter" provider, and set the model to "gpt-5.1" (the configuration can be successfully completed at this point);
3. Wait for a period of time;
4. Reopen the newly added model configuration created just now, and you will find that it shows "gpt-5.1 does not exist", and the model has been automatically switched to "Claude sonnet 4.5".

Expected result

I expect to configure these models correctly.

Actual result

The model configuration page indicates that the model does not exist in OpenRouter.

Variations tried (optional)

No response

App Version

afer v3.33.1

API Provider (optional)

OpenRouter

Model Used (optional)

gpt-5.1 series, gemini-3, and grok-4.1-fast

Roo Code Task Links (optional)

No response

Relevant logs or errors (optional)

[roomote]

I'd like to help with this issue, but I need some clarification to ensure I implement the right solution. Could you please provide more details on the following:

1. Exact model IDs: Could you provide the exact model IDs you're trying to use? For example, is it openai/gpt-5.1 or something else? The exact strings would help me reproduce the issue.

2. OpenRouter documentation: Can you confirm these models appear in OpenRouter's official model list at https://openrouter.ai/models? Some of the models mentioned (like GPT-5.1) don't appear to be publicly documented.

3. API response: When the models work initially, can you check the network tab in your browser's developer tools to see what models are returned from the OpenRouter API? This would help identify if the API is returning different results over time.

4. Consistency: Does this happen every time after ~5 minutes, or is it intermittent?

Please tag @roomote in your response with the answers, and I'll be happy to implement the fix once I have this information.

[zx2021210538]

@roomote

1. The exact model IDs include openai/gpt-5.1, google/gemini-3-pro-preview, and x-ai/grok-4.1-fast:free. (In fact, the entire gpt-5.1 series has this bug.)
2. I can confirm that these model versions do appear in the Openrouter model list, as I have checked them on the official website.
3. I don't have information about this part for the time being.
4. This bug occurs a few minutes after opening VS Code and Roo Code for the first time, and it can be reproduced every time VS Code and Roo Code are reopened.

[roomote]

Thanks for the detailed information! I can see the issue with the OpenRouter models (openai/gpt-5.1, google/gemini-3-pro-preview, and x-ai/grok-4.1-fast:free) disappearing after a few minutes. I will investigate this caching/validation issue and work on a fix. I'll open a PR to share my progress.

[roomote]

I've opened PR #9598 with a fix for this issue: #9598

The fix improves the OpenRouter model validation to be more lenient with newer models that may not be in the cached model list yet. It uses pattern matching to allow known model formats (like gpt-5.1, gemini-3-pro-preview, and grok-4.1-fast:free) even when they're not in the cache.

Please review and let me know if you have any feedback!

[mrubens]

@roomote please close that PR, it's not on the right track. My strong suspicion is that it's related to the caching we added in #9410.

[roomote]

I've seen your comment @mrubens and will investigate the caching issue related to PR #9410. I'll explore that direction and work on a proper fix.