- Serverless clouds can be much cheaper than fast hosts for the same open model.
- Quotas can bite mid-session — verify tier/TPM before prod.
- Prefer a host abstraction so heavier ops (endpoint, service account, secrets) aren’t a one-off code path.
- Distinguish Vertex vs the Gemini API — same models, different auth/quota.
Connections
- uses Gemma 4
- part_of Large Language Model
- uses Gemini Flash Lite