Case Study: LMArena accelerates LLM web development evaluations with E2B

LMArena launches WebDev Arena in 2 hours with E2B, powering 50,000+ model comparisons

The customer, LMArena, a UC Berkeley research team, faced significant technical challenges in benchmarking large language models for web development tasks. Their platform, WebDev Arena, required a way to securely and simultaneously execute code from multiple LLMs at scale, ensuring speed, isolation, and reliability for fair comparisons. They implemented E2B's secure cloud sandboxes to solve this problem.

E2B provided isolated virtual machine environments that started quickly and could run multiple LLMs' code simultaneously without interference. This solution enabled LMArena to reliably execute and compare code outputs. The results were substantial; following the implementation, the platform started over 230,000 E2B sandboxes and ran more than 50,000 model comparisons, successfully scaling to meet high user demand.

LMArena