NVIDIA JetPack 7.2 Release: Agentic AI Comes to Jetson AGX Orin
NVIDIA JetPack 7.2 is more than a routine software update. While JetPack 6.2 focused on unlocking more AI performance from Jetson Orin through Super Mode, JetPack 7.2 introduces a new software foundation designed for Agentic AI, robotics, and production-scale edge deployments.
Built on Ubuntu 24.04, Linux Kernel 6.8, and CUDA 13, JetPack 7.2 brings the latest compute stack to the entire Jetson Orin family while adding new capabilities such as Jetson Agent Skills, NemoClaw support, official Yocto integration, and improved performance for Jetson AGX Orin 32GB.
So what actually matters for developers upgrading from JetPack 6.2 to 7.2?
Jetson Is Now Ready for Agentic AI
The biggest theme of JetPack 7.2 is Agentic AI.
Rather than simply running perception models or LLM inference, NVIDIA is positioning Jetson as a platform capable of hosting AI agents that can reason, plan, and interact with the physical world.
JetPack 7.2 introduces two major additions:
- One-command deployment of NVIDIA NemoClaw
- Jetson Agent Skills for development automation, assisting with tasks such as Linux customization, memory optimization, performance tuning, and model benchmarking.
Instead of manually configuring every step, developers can leverage AI-powered workflows to accelerate deployment and reduce engineering effort.
A New Software Foundation: Ubuntu 24.04 + CUDA 13
JetPack 7.2 upgrades the entire Jetson software stack:
| JetPack 6.2 | JetPack 7.2 |
| Ubuntu 22.04 | Ubuntu 24.04 |
| Linux Kernel 5.15 | Linux Kernel 6.8 |
| CUDA 12.x | CUDA 13 |
| Jetson Linux 36.x | Jetson Linux 39.2 |
JetPack 7.2 unifies the software foundation across Jetson Orin and future Jetson Thor platforms, reducing migration effort and simplifying long-term maintenance. Developers can access the latest CUDA capabilities while building applications that are easier to scale across different Jetson generations.
During our testing with JetPack 6.2 and JetPack 7.2 on reComputer AGX Orin developer bundle, we used the same Jetson AGX Orin 32GB module and compared 2 version software environments with an LLM inference workload of Qwen 3.5 27B model. Check the whole overview of the test results on Seeed wiki guidance.
| Item | JetPack 6.2 test | JetPack 7.2 test |
| Module | Jetson AGX Orin 32GB | Jetson AGX Orin 32GB |
| Jetson Linux | L4T 36.4.3 | L4T 39.2 |
| Ubuntu | 22.04 | 24.04 |
| CUDA | 12.6 | 13.x |
| Inference engine | llama.cpp | llama.cpp |
| Model | Qwen3.5-27B-Q4_K_M.gguf | Qwen3.5-27B-Q4_K_M.gguf |
| Main parameters | -ngl 999 -fa on -ub 512 -t 12 | -ngl 999 -fa on -ub 512 -t 12 |
We’ve noticed that the JetPack 7.2 reduced memory usage after model loading and improved both prompt processing speed and token generation speed.
| Metric | JetPack 6.2 | JetPack 7.2 | Observed change |
| Memory after model load | 24.6 GB / 30 GB | 14.7 GB / 30 GB | About 40% lower |
| GPU frequency during inference | 930 MHz | 1.36 GHz | Higher boost frequency |
| Prompt processing | 18.2 tokens/s | 25.8 tokens/s | About 41.8% faster |
| Token generation | 4.3 tokens/s | 5.5 tokens/s | About 27.9% faster |
The most practical improvement is the memory headroom. In JetPack 6.2, the 27B model occupied most of the available system memory immediately after loading. In JetPack 7.2, the system preserves roughly 10 GB of additional free memory, leaving significantly more room for concurrent workloads such as vision preprocessing, robotics middleware, or other services alongside the LLM workload.
For a major-version migration such as JetPack 6.x to JetPack 7.x for Jetson firmware download, flashing, device connection, and OTA operations, you can also find Seeed Jetson DevelopTool to complete the general workflow:
- Complete first boot setup and verify the system version.
- Install and open Seeed Jetson DevelopTool.
- Select the Jetson product or carrier board model.
- Select the target L4T or JetPack version that matches JetPack 7.2.
- Download and extract the BSP package from the tool.
- Put the Jetson device into Force Recovery Mode.
- Detect the device from the host PC.
- Start flashing and wait until the device reboots.
Official Yocto Support Finally Arrives
One of the most important updates for industrial customers is official Yocto support.
Historically, many production deployments relied on community-maintained projects such as OE4T and Meta-Tegra to build customized Jetson Linux distributions.
With JetPack 7.2, NVIDIA now provides official Yocto support, including validated recipes, reference images, CI/CD maintenance, and documentation.
For companies deploying robots, smart machines, and edge AI systems at scale, Yocto offers several advantages:
- Smaller and leaner system images
- Better reproducibility
- Easier OTA management
- Long-term maintainability
While this feature may receive less attention than CUDA 13 or Agentic AI, it arguably has the greatest impact on commercial deployment readiness.
More Performance from Existing Hardware
JetPack 7.2 also delivers additional performance improvements without requiring new hardware.
The most notable update is the introduction of Super Mode for Jetson AGX Orin 32GB, increasing AI performance from 200 TOPS to 241 TOPS—more than a 20% improvement over the standard configuration.
For developers running:
- Vision-language models (VLMs)
- Robotics perception pipelines
- Multi-camera applications
- Edge AI agents
This provides additional headroom while maintaining the same hardware platform.
JetPack 7.2 also extends the JetPack 7 software stack to the entire Jetson Orin family, ensuring Orin users can benefit from future platform improvements without waiting for a new hardware generation.
JetPack 7.2 vs JetPack 6.2: The Real Difference
If JetPack 6.2 was about extracting more performance from Jetson Orin, JetPack 7.2 is about preparing Jetson for the next wave of Physical AI applications.
The headline features—Agent Skills, NemoClaw, CUDA 13, Ubuntu 24.04, and Yocto support—all point toward the same direction:
Moving Jetson from an edge inference platform to a production-ready platform for Agentic AI and Physical AI.
For developers building robots, autonomous machines, and intelligent edge systems, JetPack 7.2 is not simply another SDK release. It is the foundation for the next generation of AI-native embedded computing.
About Author
