Understand this story
On November 13, 2026, AWS released the next-generation Amazon Bedrock AgentCore runtime. According to AWS Machine Learning, each session still runs in an independent Firecracker MicroVM, with session startup speed improved by up to 15 times, compute scaled up to 4 times, and runtime efficiency improved by up to 68%; a single session supports up to 8 vCPUs and 32 GB of memory, container images up to 10 GB, and startup time under 2 seconds unaffected by image size and concurrency. The runtime reclaims memory within minutes after the agent releases it, idle sessions can scale down to zero, and it supports dynamically mounting data to file systems such as S3, keeping data within the same trust boundary. Unite.AI reported the same day that the new version switches to snapshot restore instead of repeated initialization, and introduces memory reclamation and committed baseline billing; Coinbase said that after migration, session cold starts dropped from nearly a minute to under 5 seconds, and Zuora said that in most scenarios memory usage per invocation fell by about 60%.
AI-generated from public reports
Latest development · Report
AWS AgentCore Runtime Upgrade: Sessions Expanded to 8 vCPU, Supports File System MountingDevelopments
1 confirmed developments
- Source:AWS Machine Learning
Reports and discussion timeline
Browse reports and discussion by time. Their evidence levels remain distinct.
Nov 13
Unite.AI
AWS AgentCore Runtime Upgrade: Sessions Expanded to 8 vCPU, Supports File System Mounting
On November 12, 2026, AWS expanded Amazon Bedrock AgentCore Runtime: a single session can now use up to 8 vCPU and 32 GB of memory and supports 10 GB container images, and AWS says session startup time has been reduced to under 2 seconds. The new version replaces repeated initialization with snapshot restore and introduces memory reclamation and committed baseline billing, allowing idle agents to scale to zero. Coinbase says that after migration, session cold starts dropped from nearly a minute to under 5 seconds, while Zuora says memory usage per invocation fell by about 60% in most scenarios. Sessions still run in isolated Firecracker microVMs, can dynamically mount file systems such as S3, and the feature for partitioning storage directories by session is planned to be available before re:Invent.
AWS Machine LearningFirst-party
AWS releases a new generation of AgentCore runtime, speeding up session startup by up to 15x
AWS released a new generation of the Amazon Bedrock AgentCore runtime. Each session still runs in an independent Firecracker MicroVM, with session startup speed improved by up to 15x, compute scaled up to 4x, and runtime efficiency improved by up to 68%. The new version supports up to 8 vCPU and 32 GB of memory per session, container images up to 10 GB, and startup times under 2 seconds, unaffected by image size and concurrency. The runtime also reclaims memory within minutes after the agent releases it, idle sessions can scale down to zero, and it supports dynamically mounting data to file systems such as S3, keeping data within the same trust boundary.
Heat trend for this story
- Current
- 5
- Period peak
- 20 Nov 13 03:00
- 24-hour change
- -50%
Move the pointer or select the chart to inspect hourly heat. Use the left and right arrow keys with a keyboard.