Use provider-specific test model instead of hardcoded gpt-4o-mini

Each provider type now uses an appropriate model for connection testing

Like what you see shipping?

Every one of these updates lands in your workspace automatically. Start free and watch it grow week after week.