Research
Work from Brainbase Labs on agents, evaluation and the systems around them.
Introducing vidjev: applying System One to video and live feeds
vidjev brings System One models to images and video. Zero-shot, with no training run: a drone that follows a car through a city, and CCTV monitoring level with a detector trained on the benchmark.
Joshua Okolo, Gokhan Egri
Choosing a System One model for your workflow
We compiled τ²-bench customer-service agents into workflows and sent their typed decisions to fast System One models. The best model per domain cut cost per conversation by 56 to 68% against the agent.
Joshua Okolo, Gokhan Egri