I've built a browser agent harness that outperforms Browser Code on their own benchmark (BU Bench v1), on success rate, speed and cost. Browser Agent: 88% success rate, $5.37, 32,694 seconds Browser Code: 78% success rate, $8.34, 47,970 seconds For browser agents to become ubiquitous, they need to be faster, cheaper, and more reliable. The harness was built with token efficiency in mind (and uses 91% fewer tokens than Browser Code) to reduce the compute and VRAM budget necessary to fine tune a small, specialised model. Because browser agent inference is intermittent, a small specialised model could serve each step with minimal GPU time. This makes per-task pricing viable, offering users cheaper, more predictable costs and giving the operator a counter-positioning advantage over frontier labs pricing their models by token usage. More about the comparison between Browser Agent and Browser Code here: https://www.pierrebarreau.com/blog/improving-the-state-of-th... If anyone is interested in fine tuning that model, would love to talk! |