Case Study: Microsoft achieves reliable computer-use model evaluation with Browserbase

Microsoft benchmarks Fara-7B on 595 WebVoyager tasks with Browserbase

Microsoft, a leading technology company, was developing Fara-7B, a new open-source computer-use model, and faced the challenge of training and evaluating it at scale. This required reliable access to real websites, high concurrency, and consistent execution environments to ensure accurate benchmarking, which is why they partnered exclusively with Browserbase.

Browserbase provided its deterministic browser infrastructure to host and independently evaluate the model. The solution ensured consistent environments and incorporated human verification. The results showed Fara-7B significantly outperformed similar open-source models on the 595-task WebVoyager benchmark, achieving new state-of-the-art results for its size and enabling new, cost-effective workflows for browser agents.

Microsoft