Perceptron AI Launches Multilook API

Perceptron AI announced on X that its Multilook API is now live. The company, which builds promptable vision models, stated that the API prefills context once per call instead of once per question. One video input can produce up to 16 answers at 32 percent of the input cost of sending queries separately, with reported median latency reductions between 2 times and 4.9 times under load. The post included a split-screen UI demo video. CEO Armen Aghajanyan retweeted the announcement from the company's account.