Case Study: LMArena scales agent evaluations with E2B sandbox infrastructure

LMArena scales to 600,000 sandboxes a day with E2B

The customer, Arena, faced the challenge of securely isolating thousands of concurrent AI agent evaluations on its large-scale platform. They needed an infrastructure solution that could instantly provision isolated environments, handle unpredictable traffic surges from new model releases, and preserve state for long-running tasks, all without diverting engineering resources from their core product. Arena selected E2B to provide their secure cloud sandbox service.

E2B implemented its sandbox platform, providing Arena with isolated microVM environments for each user session. This solution delivered secure, ready-to-use workspaces and enabled the preservation of state between sessions. The results were significant, with E2B supporting approximately 600,000 sandboxes per day and scaling to provide thousands of additional concurrent sandboxes within hours for a major launch. This allowed Arena to handle massive traffic surges and keep its engineering focus on its evaluation platform.

LMArena