Nvidia
Nvidia packs five vision tasks into one 3B model — runs on a single GPU
Nvidia's LocateAnything is a 3B vision model on a single GPU that unifies five visual detection tasks. It bucks "bigger is better"—realistic for mid-s
3 articles tagged with this topic
Nvidia's LocateAnything is a 3B vision model on a single GPU that unifies five visual detection tasks. It bucks "bigger is better"—realistic for mid-s
DeepSeek's new vision model reads text from any photo—receipts, contracts, product shots. ¥0.1–0.3 per image, WeChat pay. I tested it on my ledger at
Liquid AI ships LFM2.5-VL-3B: 2GB, 3.1B params, runs locally on iPhone 17. ScreenSpot-v2 score jumped 6 to 78.7 — local models now truly see.