Three New Models Join SageMaker JumpStart: LocateAnything, Qwen-AgentWorld, and Qwen3.5!

Hey, it's me! Today three new models joined SageMaker JumpStart all at once.

AWS What's New aws.amazon.com

What was announced?

According to AWS's What's New, three new foundation models were added to Amazon SageMaker JumpStart: NVIDIA's "LocateAnything-3B," Qwen's "Qwen-AgentWorld-35B-A3B," and Qwen's "Qwen3.5-122B-A10B." Each one specializes in something different.

Why it matters

Instead of one generalist model trying to do everything, these three are each specialized: visual object localization, agent environment simulation, and large-scale multimodal reasoning. That's great news if you want to pick a model that fits your specific enterprise use case instead of settling for a jack-of-all-trades.

What changes

You can deploy any of them right away from the SageMaker JumpStart model catalog or the Python SDK. Here's what each one brings:

• LocateAnything-3B (NVIDIA): fast, high-quality visual grounding — locating objects from natural language instructions
• Qwen-AgentWorld-35B-A3B (Qwen): a "language world model" that predicts agent behavior and environment states
• Qwen3.5-122B-A10B (Qwen): a large sparse MoE model built for high-performance multimodal reasoning

Dive Deep

Let's go a bit deeper on each one.

LocateAnything-3B uses a "Parallel Box Decoding (PBD)" framework that decodes bounding boxes and points as atomic units in a single step, preserving geometric coherence while unlocking substantial parallelism. It's built for precise object localization, dense detection, and point-based localization across enterprise intelligence and physical AI applications.

Qwen-AgentWorld-35B-A3B is the first language world model to cover all seven interaction domains — tool calling, search, terminal, software engineering, Android, web, and OS interaction — in a single model. It predicts the next environment state given an agent's action and interaction history via long chain-of-thought reasoning, trained on over 10 million real-world interaction trajectories.

Qwen3.5-122B-A10B has 122B total parameters but activates only 10B per token, using a sparse MoE design with 256 experts and a 262K context window — aiming for strong reasoning performance while staying efficient.

Wrap-up

• LocateAnything-3B, Qwen-AgentWorld-35B-A3B, and Qwen3.5-122B-A10B all joined Amazon SageMaker JumpStart
• LocateAnything-3B: a lightweight model specialized in visual object localization
• Qwen-AgentWorld-35B-A3B: a language world model simulating agent environments across seven domains
• Qwen3.5-122B-A10B: a sparse MoE model with 256 experts and a 262K context window

If you need to pick a model for visual recognition, agent simulation, or large-scale multimodal reasoning, this update is for you!