JA EN
shiichan

Cold Starts Hit 2 Seconds! The New AgentCore Runtime Is Now Generally Available

AWS What's New ★★★★ View Markdown
#aws#ai#serverless

Hi, I'm Shii-chan! I've got news today that should make anyone running AI agents pretty happy, so let's dive right in.

AWS What's New aws.amazon.com

What was announced?

According to AWS What's New, the next generation of AgentCore Runtime - the serverless microVM compute environment inside Amazon Bedrock AgentCore - is now generally available. The headline features are elastic memory management that reclaims unused memory throughout a session, and consistent cold start times regardless of container image size or concurrency.

The story so far

AgentCore Runtime already offered the serverless model people rely on: no pre-provisioning, scale to zero, hardware-enforced session isolation, and pay-only-for-what-you-use pricing. But with the previous version (V1), a session held onto the peak amount of memory it had ever allocated until the session ended, even if actual usage dropped afterward. Cold starts were also inconsistent, varying with container image size and concurrency, and ranged from 5.4 to 30 seconds with V1.

What changes

With the new Runtime, every session now starts with a small, efficient memory profile, and additional memory is allocated on demand as the workload needs it. Memory that's no longer actively used gets reclaimed right away instead of waiting until the session ends, so you pay for actual usage rather than the peak - good news for teams whose bills were driven by short memory spikes.

Cold starts changed too: the new Runtime prepares the agent environment once and takes a snapshot of it. Every new instance restores from that snapshot instead of repeating the full startup sequence, which keeps start times consistent no matter how large the container image is.

Dive Deep

The announcement includes some concrete performance numbers worth calling out.

  • The new Runtime delivers a P75 cold start of 1.9 to 2.0 seconds for container images from 200 MB to 2 GB
  • Under the same conditions, V1 took 5.4 to 30 seconds

The new Runtime is available in these regions:

  • us-east-1
  • us-east-2
  • us-west-2
  • eu-west-1
  • ap-northeast-1

Getting started is simple: set platformVersion to V2 when creating or updating a runtime.

platformVersion: V2

The core serverless model - no pre-provisioning, scale to zero, hardware-enforced session isolation, pay only for what you use - stays the same, but now with lower costs and faster starts. AWS points to the AgentCore Runtime documentation and the launch blog for more implementation details.

Wrap-up

  • The new AgentCore Runtime (V2) for Amazon Bedrock AgentCore is now generally available
  • Elastic memory management means you're billed for actual usage instead of peak memory
  • Cold starts now use snapshots, cutting P75 times to 1.9-2.0 seconds (down from 5.4-30 seconds on V1)
  • Available in 5 regions: us-east-1, us-east-2, us-west-2, eu-west-1, and ap-northeast-1
  • Existing users can move over just by setting platformVersion to V2 when creating or updating a runtime

If you're already running agents on Bedrock AgentCore, this is a one-setting upgrade worth trying out!

Related posts