What LLM model should I use for Browser Use? The Definitive Browser AI Benchmark

A statistically rigorous benchmark compares the speed, cost, and accuracy of frontier large language models for browser automation. The proprietary `bu-2` model emerges as the top performer with the best performance and lowest latency. For standalone options, Gemini is recommended as the best value for general web browsing, while Claude is unparalleled for tasks involving custom code execution or complex data extraction. The analysis also notes that recent OpenAI models have regressed on web tasks and open-source models are not yet reliable for complex automation.