Google adds Flex and Priority tiers to the Gemini API for cost and reliability control

On Apr 02, 2026, Google introduced two new service tiers in the Gemini API: Flex and Priority. The company is responding to a common agent design problem, where developers want cheaper handling for background work but stronger reliability for user-facing requests that cannot tolerate interruptions during peak demand.

Google's argument is architectural as much as commercial. Until now, teams often had to split background logic across standard synchronous serving and the asynchronous Batch API. Google says Flex and Priority let developers keep both background and interactive traffic on standard synchronous endpoints, then control behavior by setting the service_tier parameter per request.

Flex Inference is the cost-optimized option. Google says it is built for latency-tolerant workloads without batch-processing overhead and delivers 50% price savings compared with the Standard API. The company highlights background CRM updates, large-scale research simulations, and agentic workflows where a model browses or thinks in the background as example use cases. Flex is available for all paid tiers and works on GenerateContent and Interactions API requests.

Priority Inference is the premium path for critical applications. Google says the tier gives requests the highest criticality so important traffic is not preempted during peak load. If usage exceeds Priority limits, overflow requests are automatically served at the Standard tier instead of failing outright. Priority is available to Tier 2 / 3 paid projects across GenerateContent and Interactions API endpoints.

• Flex lowers inference cost while keeping a synchronous developer experience.
• Priority increases assurance for time-sensitive traffic and adds graceful downgrade behavior.
• Together, the tiers make request-level economics and reliability part of application design.

The strategic implication is that model APIs are evolving into traffic-management layers for agentic applications. Google is not only selling tokens; it is selling differentiated runtime behavior that maps to specific business workloads.