Gemini 3.8 makes the choice between fast dialogue and deeper deliberation explicit, while Gemini 3.7 offers thinking configurations that trade off different needs. OpenRouter’s routing across hundreds of models puts that choice across providers as well. Builders can treat latency, reasoning depth, and token economics as runtime decisions rather than fixed model selections.
From the Field Notes:
- A Model Launch Is Now a Routing and Distribution Event (2026-09-20)
- Gemini 3.8 turns voice-agent latency into a routing policy (2026-09-20)
- Stripe agrees to acquire OpenRouter (2026-09-18)