AWS Machine Learning Blog · Agent Runtime Update

Next generation Amazon Bedrock AgentCore runtime

AWS has updated the AgentCore runtime, the managed compute layer for Amazon Bedrock AgentCore, where each agent session runs in its own hardware-isolated Firecracker microVM. Sessions now scale to 8 vCPUs and 32 GB, support container images up to 10 GB, and start in under 2 seconds without warm pools. The update also reclaims memory within minutes of release, scales idle agents to zero, and adds committed baseline pricing for always-on agents. AWS reports up to 15x faster starts, 4x more compute and up to 68% more efficient runs, though the page text is truncated and availability details are not in the excerpt.