Amazon Bedrock Service Tiers

Amazon Bedrock now offers two new service tiers in addition to the existing Standard tier:

• Priority - Premium tier offering faster response times and prioritized compute for time-sensitive applications
• Flex - A cost-effective option for non-time-critical workloads where requests may see increased latency

The Standard tier remains available as the default service option with reliable performance for everyday AI applications. The new service tiers provide additional inference options depending on performance and cost requirements.

The Priority tier is a premium service tier that provides faster response times and preferential processing for time-sensitive workloads. For most models that support Priority tier, customers can realize up to 25% better output tokens per second (OTPS) latency compared to standard tier.

Priority is a good fit for:

• Real-time customer interactions
• Interactive user experiences like real-time language translation
• Mission-critical applications requiring immediate responses

Priority tier requests receive access to more compute resources and processing priority over other tiers, providing faster performance even when the system is under heavy load.

The Flex tier is a cost-effective service tier designed for non-time-critical workloads. It can be used in cases where your applications and agentic workflows can tolerate some increase in latency.

The Flex tier is ideal for:

• Model evaluations
• Content summarization, labeling and annotation
• Multi-step agentic workflows

Flex may experience longer latencies than standard tier. During high traffic, Flex tier requests are processed after Standard tier requests. It is designed for non-interactive workloads that can tolerate these longer latencies.

While both options are cost-effective for non-time-critical workloads, they serve different use cases:

• Flex: Real-time single API call suitable for applications that can tolerate some increase in latency but still need synchronous processing.
• Batch: Asynchronous processing of large datasets where you submit multiple prompts at once and retrieve results later from Amazon S3. Some examples of typical use cases for batch inference include: create large volumes of marketing content, document classification, or data extraction.

We have several models available from leading providers like OpenAI, DeepSeek, Qwen and Amazon.

For a complete list check the documentation.

Consider these factors:

• Priority: Choose if you need faster response times for mission critical applications.
• Standard: Choose for everyday AI applications that need reliable performance.
• Flex: Choose if cost optimization is your priority and your application can tolerate increased latency.

Yes, you can choose different service tiers based on your specific use case and requirements for each API call or application.

Yes, each service tier has a different pricing structure. Flex tier offers cost savings over standard pricing for flexible workloads, while Priority tier commands a premium for faster performance and prioritized compute. Please see the Amazon pricing page for specific prices by model.

The new service tiers are available through the Amazon Bedrock console and API. You can use the same inference API as the default standard tier by just adding an extra parameter to select service tier at invoke time. Please review documentation for more details.

Existing applications will continue to work with the Standard tier. To take advantage of the new service tiers, you'll need to update your API calls to specify the desired service tier for supported models.

For detailed pricing information and the latest updates on model availability, please refer to the Amazon Bedrock pricing page or contact your AWS account team.