Hey, thanks for the detailed report.
What you’re seeing is actually by design. For custom OpenAI-compatible models that aren’t in our catalog, we set 1M as a safe upper bound so we don’t accidentally set the limit too low for a model that really supports a large context. We don’t currently have auto-detection of the context window from the provider, and there also isn’t a UI yet to configure the limit for a custom BYOK model.
So it’s not a bug, but the request to let users set the context window for custom models manually, or at least use a more conservative default, is valid. There’s already a similar feature request here: Unlock Full Context Window with Own API Keys It’s worth commenting there to increase visibility.
I can’t give an ETA for when this option will be available. If there are updates, We’ll follow up.