AI Briefing
KO

A new way to control cost and reliability in the Gemini API

·2026.04.03 01:00

Key point

Flex and Priority have been added to the Gemini API, letting you adjust cost and reliability by tier.

Details

Two new service tiers, Flex and Priority, have been added to the Gemini API. From a single unified interface, you can finely tune cost and reliability according to the importance of a request.

As AI expands beyond simple chat into complex agent workflows, developers have typically had to handle two kinds of tasks separately. Large volumes of background task don't need an immediate response, while interactive task requires high reliability and fast response. Until now, developers had to split usage between standard synchronous serving and the Batch API to accommodate this difference, but the new tiers bridge that gap.

Flex Inference is a cost-optimized tier for tasks that can tolerate latency.

  • 50% price reduction: available at half the price compared to the Standard API.
  • Synchronous calls: no need to manage input/output files or poll for job completion.
  • Suitable use cases: background CRM updates, large-scale research simulations, agentic workflows that browse or think in the background.

Flex can be used by simply setting the service_tier parameter, and it is available for the entire paid tier and for GenerateContent and Interactions API requests.

Priority Inference is the highest-reliability tier for critical traffic. It focuses on guaranteeing that requests are not preempted even during peak hours.

  • Top-level criticality: provides higher reliability for critical requests.
  • Automatic graceful downgrade: if the Priority limit is exceeded, requests are automatically handled by the Standard tier instead of failing.
  • Response transparency: you can check which tier handled the request in the API response.
  • Suitable use cases: real-time customer support bots, live content moderation pipelines, time-sensitive requests.

Priority can also be used by specifying service_tier. This tier is available for the GenerateContent API and Interactions API of Tier 2 / 3 paid projects, and detailed pricing can be found in the Gemini API documentation and cookbook examples.

This summary was generated automatically by AI. Check the original for the author's claims and context. Copyright belongs to the original author.

Our guide explains how the AI works. Report summary errors, attribution issues, or removal requests via Contact.