Not much. Running gpt-oss:20b via Ollama completely offline. For personal stuff I never use remote models.