A 1.3B model just shipped that runs on your phone, and the labs obsessed with frontier scores won't see this story coming

The article discusses the release of OpenBMB's MiniCPM-V 4.6, a 1.3B-parameter multimodal model that can run on phones. This model is notable for its focus on deployment and usability, rather than benchmark scores. The author believes that this shift towards local-first multimodal AI will be the next big consumer story, and that labs focused on frontier scores will miss it. The model's ability to provide fast, private, and reliable results will be more valuable to most users than high accuracy at the cost of network latency and cost.