API Gemini: Flex and Priority Redefine Cost and Reliability

Introduction of Flex and Priority in the Gemini API

The Gemini API introduces two new service options: Flex and Priority. These additions aim to provide users with more precise control over the costs and reliability of their operations, all through a unified and simplified interface.

Flex Inference: An Economical Solution

The Flex Inference tier is designed to reduce costs while accepting some latency. It targets workloads where response time is not critical, allowing for substantial savings.

• 50% Cost Reduction: By opting for Flex, users can cut their costs in half compared to the Standard API. This is achievable by accepting lower reliability and increased latency.

• Simplified Synchronous Interface: Unlike the Batch API, Flex uses a synchronous interface, eliminating the need to manage input and output files or monitor task completion.

• Ideal Applications: This tier is perfect for background CRM updates, large-scale research simulations, and workflows where the model operates in the background.

To get started, simply configure the service_tier parameter in the request. The Flex tier is available for all paid tiers and accessible via the GenerateContent and Interactions API requests.

Priority Inference: Reliability for Critical Applications

The Priority Inference tier is designed to offer maximum reliability at a premium cost, ensuring that critical requests are not interrupted, even during high platform usage.

• Maximum Criticality Assurance: Requests processed in Priority mode benefit from increased reliability, even during peak loads.

• Overage Management: In the event of exceeding priority limits, excess requests are automatically downgraded to Standard level, preventing failures and ensuring operational continuity.

• Complete Visibility: The API provides information on the service level used for each request, offering total transparency on performance and billing.

• Recommended Uses: This tier is ideal for real-time customer support bots, live content moderation pipelines, and time-sensitive applications.

To use Priority Inference, simply set the service_tier parameter accordingly. This tier is available for users with paid projects at level 2 or 3 on the GenerateContent and Interactions API endpoints.

For more details on pricing and to optimize your production levels, consult the Gemini API documentation. Executable code examples are available in the recipe book to help you get started.