Tag
#vision language models
Your camera now understands sentences: a 2026 field guide to vision AI models
Vision AI has split into four families: frontier VLMs that can now draw accurate boxes, promptable segmenters like SAM 3, real-time edge detectors like YOLO26 and RF-DETR, and label-free backbones like DINOv3. Here is what each one is good at, with real 2026 benchmarks, and how to pick.
Robots that learn a task from one demonstration: inside 2026's vision-language-action models
In one week of November 2026, Google DeepMind and NVIDIA released robot foundation models that learn from human video and a handful of demonstrations. How vision-language-action models work, what a humanoid actually did on a Mercedes-Benz line, and what is still hard.