Eagle

Eagle: Frontier Vision-Language Models with Data-Centric Strategies

What is Eagle?

Eagle: Frontier Vision-Language Models with Data-Centric Strategies

Overview of Eagle

What it is

Eagle is a family of frontier vision-language models (VLMs) developed by NVIDIA. It explores data-centric strategies across general-purpose multimodal understanding, long-context reasoning, and embodied applications. The Eagle family includes models such as VFlash, LocateAnything, Eagle 2.5, Eagle 2, and the original Eagle. These models have been used as a research and development platform supporting multiple NVIDIA efforts, including Llama-Nemotron Nano VLM, Nemotron VLMs, NeMo Retriever, NVIDIA Isaac GR00T N1/N1.5/N1.6, and Cosmos.

Key features

  • VFlash: A generalist decoding framework for fast, accurate visual planning in Physical AI. It uses Flash Primitive Decoding to generate complete visual primitives (trajectories, boxes, points, dense 3D motion) in one parallel drafter pass.
  • LocateAnything: A generalist vision-language grounding model that performs diverse localization tasks under a unified VLM, including document understanding, GUI grounding, dense object detection, and OCR. It uses Parallel Box Decoding for faster throughput.
  • Eagle 2.5: A frontier VLM with state-of-the-art image and video understanding, focusing on framework and data strategy for long-context multimodal understanding.
  • Eagle 2: A frontier VLM exploring post-training data strategies for image understanding.
  • Eagle: The original model exploring the design space for vision-centric VLMs with mixture-of-encoders.

Use cases

Eagle models are designed for general-purpose multimodal understanding, long-context reasoning, and embodied applications. Specific use cases include robot and driving trajectory planning, dense object detection, document understanding, GUI grounding, OCR, video analysis and captioning, and visual planning for Physical AI. The models have been adopted as VLM backbones in NVIDIA's GR00T N1/N1.5/N1.6 and other enterprise and Physical AI systems.

Getting started

See the repository README for getting started guides for VFlash, LocateAnything, Eagle 2.5, and Eagle.

License and pricing

The license is Apache-2.0. The project is open source.

Based on the project's public README and repository metadata.

Add this badge to your README

Featured on YunoDB Badge styles & help

Maintain this project? Show that it is featured on YunoDB.

Markdown[![Featured on YunoDB](https://yunodb.com/badge/nvlabseagle.svg)](https://yunodb.com/tools/nvlabseagle?utm_source=github&utm_medium=badge)
HTML<a href="https://yunodb.com/tools/nvlabseagle?utm_source=github&utm_medium=badge"><img src="https://yunodb.com/badge/nvlabseagle.svg" alt="Featured on YunoDB" height="20"></a>