Beam V1 was deprecated on September 16, 2024, and all V1 data and
deployments have been deleted.This page is kept as a reference for users migrating old V1 code to V2.
What is V2?
V2 is the next-generation version of Beam.
How is V1 different from V2?
From a user perspective, it’s different in a few ways:
What are the benefits of using V2?
Most importantly, V2 is the future of Beam – all our new functionality is being built into V2, and this is the future of the platform.
Why did you decide to make a new dashboard, CLI, etc?
We have grown a lot over the past 12 months, and we ran into problems that prevented us from scaling to support our new users.
In addition, there were other gaps in V1 that we’ve closed in V2:
- Not enough GPUs – very few A100s, and only had T4s and A10Gs
- Using outputs required you to write callback logic
- No 1:1 mapping between dashboard and CLI (can’t create secrets or stop containers in CLI, etc.)
- Very specific to inference and not really optimized for other workflows like data processing or model training
What about reliability?
V2 is already powering production workloads for our largest customers. We have run load tests on all customer apps before deploying them to production in V2. Reliability remains our highest priority to you as your infrastructure provider.
How do I migrate?
V1 is no longer available, so old V1 apps need to be refactored against the V2 SDK. The Quickstart covers the new workflow, and you can message us on Slack if you have questions while migrating your code.
It sounds like a lot of stuff is changing. What isn’t changing?
- Fast cold starts
- Reliability
- Developer experience
- White-glove customer support whenever you have questions about using Beam
