Amazon SageMaker JumpStart broadened its foundation model catalog with LocateAnything-3B focused on precise object localization from language instructions using a parallel box decoding method, and two Qwen models addressing agent environment simulation and large-scale multimodal reasoning. LocateAnything supports diverse applications in enterprise intelligence and physical AI by enhancing rapid, accurate visual grounding.
Qwen-AgentWorld-35B-A3B simulates agent interactions across multiple domains including software and OS tasks, enabling predictive environment state modeling. Qwen3.5-122B-A10B combines Gated Delta Networks and sparse Mixture-of-Experts architecture to deliver efficient large-scale multimodal reasoning with a 262K token context window, suitable for demanding AI use cases including coding and visual understanding.
Source & further reading
LocateAnything-3B, Qwen-AgentWorld-35B-A3B, and Qwen3.5-122B-A10B models now available on Amazon SageMaker JumpStart ↗Source published 2026-08-11. This report was prepared with AI assistance from the linked primary-source material. Claims and limitations should be checked against the original announcement or research.