AWS Lambda Expands to Four Execution Models, Supporting Workflows Up to One Year and Code Sandboxes
Key point
AWS Lambda has expanded to a total of four execution models by adding three new options—always-on, long-running workflows, and code sandboxes—alongside the existing per-invocation execution model.
Details
As of August 2026, AWS Lambda offers four execution models: Lambda Functions, Lambda Managed Instances, Lambda Durable Functions, and Lambda MicroVMs. While the existing per-invocation execution model, Lambda Functions, maintains a 15-minute limit, the new models extend execution duration up to one year and offer diverse isolation levels and state management approaches.
Lambda Managed Instances runs code on customer-owned EC2 instances, keeping the execution environment always active without freezing. It can handle multiple invocations concurrently within a single environment, making it suitable for parallel processing workloads; however, since isolation between invocations is removed, care must be taken with thread safety and state management. Billing consists of the sum of request fees, EC2 instance fees, and management fees.
Lambda Durable Functions supports durable execution via checkpoint and replay mechanisms, allowing execution to resume from the point of interruption. A single execution can last up to one year, making it useful for multi-step workflows or AI orchestration. It operates as an execution mode on top of basic Lambda functions and requires no separate infrastructure.
Lambda MicroVMs exposes Firecracker directly to provide VM-level isolation. Accessed via dedicated HTTPS endpoints, it is designed as a sandbox for safely executing user- or AI-generated code. It runs for up to 8 hours and maintains state through memory and disk snapshots.
Model selection depends on the nature of the workload. Basic Functions are optimized for event-based one-off processing, Managed Instances for always-on parallel processing, Durable Functions for long-running workflows, and MicroVMs for code sandboxes.
This summary was generated automatically by AI. Check the original for the author's claims and context. Copyright belongs to the original author.
Our guide explains how the AI works. Report summary errors, attribution issues, or removal requests via Contact.