FLUX 3 Action
VLMBlack Forest Labs' world action model for robotics, with a guide to generating action predictions on Jetson Thor using LeRobot.
Supported Models
Run the latest generative AI models on your Jetson device
Featured Models
Latest releases with day-0 support on Jetson
All Models
Browse by family cards or switch to a sortable table.
Directory view
NVIDIA Nemotron
Nemotron 3.5 Lightning
Nemotron 3.5 Lightning
Quick Start Runner
Inference Engine
Loading command...Commands are auto-generated based on your configuration settings.
Advanced configuration
Authentication
This model requires a Hugging Face access token. The token is inserted into the command and never stored.
vLLM Configuration
Configure vLLM server parameters. Leave empty to use defaults.
Maximum context length the model can handle
Fraction of GPU memory to use (0.1 - 1.0)
Nemotron3 Nano 4B
Nemotron3 Nano 4B
Quick Start Runner
Inference Engine
Loading command...Commands are auto-generated based on your configuration settings.
Advanced configuration
Authentication
This model requires a Hugging Face access token. The token is inserted into the command and never stored.
vLLM Configuration
Configure vLLM server parameters. Leave empty to use defaults.
Maximum context length the model can handle
Fraction of GPU memory to use (0.1 - 1.0)
Nemotron3 Nano 30B-A3B
Nemotron3 Nano 30B-A3B
Quick Start Runner
Inference Engine
Loading command...Commands are auto-generated based on your configuration settings.
Advanced configuration
Authentication
This model requires a Hugging Face access token. The token is inserted into the command and never stored.
vLLM Configuration
Configure vLLM server parameters. Leave empty to use defaults.
Maximum context length the model can handle
Fraction of GPU memory to use (0.1 - 1.0)
Nemotron Nano 9B v2
Nemotron Nano 9B v2
Quick Start Runner
Inference Engine
Loading command...Commands are auto-generated based on your configuration settings.
Advanced configuration
Authentication
This model requires a Hugging Face access token. The token is inserted into the command and never stored.
vLLM Configuration
Configure vLLM server parameters. Leave empty to use defaults.
Maximum context length the model can handle
Fraction of GPU memory to use (0.1 - 1.0)
Nemotron 3 Super 120B-A12B
Nemotron 3 Super 120B-A12B
Quick Start Runner
Inference Engine
Loading command...Commands are auto-generated based on your configuration settings.
Advanced configuration
Authentication
This model requires a Hugging Face access token. The token is inserted into the command and never stored.
vLLM Configuration
Configure vLLM server parameters. Leave empty to use defaults.
Maximum context length the model can handle
Fraction of GPU memory to use (0.1 - 1.0)
Nemotron Nano 12B VL
Nemotron Nano 12B VL
Quick Start Runner
Inference Engine
Loading command...Commands are auto-generated based on your configuration settings.
Advanced configuration
Authentication
This model requires a Hugging Face access token. The token is inserted into the command and never stored.
vLLM Configuration
Configure vLLM server parameters. Leave empty to use defaults.
Maximum context length the model can handle
Fraction of GPU memory to use (0.1 - 1.0)
Nemotron 3 Nano Omni
Nemotron 3 Nano Omni
Quick Start Runner
Inference Engine
Loading command...Commands are auto-generated based on your configuration settings.
Advanced configuration
Authentication
This model requires a Hugging Face access token. The token is inserted into the command and never stored.
vLLM Configuration
Configure vLLM server parameters. Leave empty to use defaults.
Maximum context length the model can handle
Fraction of GPU memory to use (0.1 - 1.0)
Meta Muse
Muse Glimmer 30B
Muse Glimmer 30B
Quick Start Runner
Inference Engine
Loading command...Commands are auto-generated based on your configuration settings.
Advanced configuration
Authentication
This model requires a Hugging Face access token. The token is inserted into the command and never stored.
vLLM Configuration
Configure vLLM server parameters. Leave empty to use defaults.
Maximum context length the model can handle
Fraction of GPU memory to use (0.1 - 1.0)
Alibaba Qwen3.6
Qwen3.6 35B-A3B (MoE)
Qwen3.6 35B-A3B (MoE)
Quick Start Runner
Inference Engine
Loading command...Commands are auto-generated based on your configuration settings.
Advanced configuration
Authentication
This model requires a Hugging Face access token. The token is inserted into the command and never stored.
vLLM Configuration
Configure vLLM server parameters. Leave empty to use defaults.
Maximum context length the model can handle
Fraction of GPU memory to use (0.1 - 1.0)
Qwen3.6 27B
Qwen3.6 27B
Quick Start Runner
Inference Engine
Loading command...Commands are auto-generated based on your configuration settings.
Advanced configuration
Authentication
This model requires a Hugging Face access token. The token is inserted into the command and never stored.
vLLM Configuration
Configure vLLM server parameters. Leave empty to use defaults.
Maximum context length the model can handle
Fraction of GPU memory to use (0.1 - 1.0)
Google Gemma4
Gemma 4 E2B
Gemma 4 E2B
Quick Start Runner
Inference Engine
Loading command...Commands are auto-generated based on your configuration settings.
Advanced configuration
Authentication
This model requires a Hugging Face access token. The token is inserted into the command and never stored.
vLLM Configuration
Configure vLLM server parameters. Leave empty to use defaults.
Maximum context length the model can handle
Fraction of GPU memory to use (0.1 - 1.0)
Gemma 4 12B
Gemma 4 12B
Quick Start Runner
Inference Engine
Loading command...Commands are auto-generated based on your configuration settings.
Advanced configuration
Authentication
This model requires a Hugging Face access token. The token is inserted into the command and never stored.
vLLM Configuration
Configure vLLM server parameters. Leave empty to use defaults.
Maximum context length the model can handle
Fraction of GPU memory to use (0.1 - 1.0)
Gemma 4 E4B
Gemma 4 E4B
Quick Start Runner
Inference Engine
Loading command...Commands are auto-generated based on your configuration settings.
Advanced configuration
Authentication
This model requires a Hugging Face access token. The token is inserted into the command and never stored.
vLLM Configuration
Configure vLLM server parameters. Leave empty to use defaults.
Maximum context length the model can handle
Fraction of GPU memory to use (0.1 - 1.0)
Gemma 4 26B-A4B
Gemma 4 26B-A4B
Quick Start Runner
Inference Engine
Loading command...Commands are auto-generated based on your configuration settings.
Advanced configuration
Authentication
This model requires a Hugging Face access token. The token is inserted into the command and never stored.
vLLM Configuration
Configure vLLM server parameters. Leave empty to use defaults.
Maximum context length the model can handle
Fraction of GPU memory to use (0.1 - 1.0)
Gemma 4 31B
Gemma 4 31B
Quick Start Runner
Inference Engine
Loading command...Commands are auto-generated based on your configuration settings.
Advanced configuration
Authentication
This model requires a Hugging Face access token. The token is inserted into the command and never stored.
vLLM Configuration
Configure vLLM server parameters. Leave empty to use defaults.
Maximum context length the model can handle
Fraction of GPU memory to use (0.1 - 1.0)
DiffusionGemma 26B-A4B
DiffusionGemma 26B-A4B
Quick Start Runner
Inference Engine
Loading command...Commands are auto-generated based on your configuration settings.
Advanced configuration
Authentication
This model requires a Hugging Face access token. The token is inserted into the command and never stored.
vLLM Configuration
Configure vLLM server parameters. Leave empty to use defaults.
Maximum context length the model can handle
Fraction of GPU memory to use (0.1 - 1.0)
NVIDIA Cosmos Reason
Cosmos Reason 1 7B
Cosmos Reason 1 7B
Quick Start Runner
Inference Engine
Loading command...Commands are auto-generated based on your configuration settings.
Advanced configuration
Authentication
This model requires a Hugging Face access token. The token is inserted into the command and never stored.
vLLM Configuration
Configure vLLM server parameters. Leave empty to use defaults.
Maximum context length the model can handle
Fraction of GPU memory to use (0.1 - 1.0)
Cosmos Reason 2 2B
Cosmos Reason 2 8B
Alibaba Qwen3.5
Qwen3.5 35B-A3B (MoE)
Qwen3.5 35B-A3B (MoE)
Quick Start Runner
Inference Engine
Loading command...Commands are auto-generated based on your configuration settings.
Advanced configuration
Authentication
This model requires a Hugging Face access token. The token is inserted into the command and never stored.
vLLM Configuration
Configure vLLM server parameters. Leave empty to use defaults.
Maximum context length the model can handle
Fraction of GPU memory to use (0.1 - 1.0)
Qwen3.5 27B
Qwen3.5 27B
Quick Start Runner
Inference Engine
Loading command...Commands are auto-generated based on your configuration settings.
Advanced configuration
Authentication
This model requires a Hugging Face access token. The token is inserted into the command and never stored.
vLLM Configuration
Configure vLLM server parameters. Leave empty to use defaults.
Maximum context length the model can handle
Fraction of GPU memory to use (0.1 - 1.0)
Qwen3.5 9B
Qwen3.5 9B
Quick Start Runner
Inference Engine
Loading command...Commands are auto-generated based on your configuration settings.
Advanced configuration
Authentication
This model requires a Hugging Face access token. The token is inserted into the command and never stored.
vLLM Configuration
Configure vLLM server parameters. Leave empty to use defaults.
Maximum context length the model can handle
Fraction of GPU memory to use (0.1 - 1.0)
Qwen3.5 4B
Qwen3.5 4B
Quick Start Runner
Inference Engine
Loading command...Commands are auto-generated based on your configuration settings.
Advanced configuration
Authentication
This model requires a Hugging Face access token. The token is inserted into the command and never stored.
vLLM Configuration
Configure vLLM server parameters. Leave empty to use defaults.
Maximum context length the model can handle
Fraction of GPU memory to use (0.1 - 1.0)
Qwen3.5 0.8B
Qwen3.5 0.8B
Quick Start Runner
Inference Engine
Loading command...Commands are auto-generated based on your configuration settings.
Advanced configuration
Authentication
This model requires a Hugging Face access token. The token is inserted into the command and never stored.
vLLM Configuration
Configure vLLM server parameters. Leave empty to use defaults.
Maximum context length the model can handle
Fraction of GPU memory to use (0.1 - 1.0)
OpenAI GPT OSS
GPT OSS 20B
GPT OSS 20B
Quick Start Runner
Inference Engine
Loading command...Commands are auto-generated based on your configuration settings.
Advanced configuration
Authentication
This model requires a Hugging Face access token. The token is inserted into the command and never stored.
vLLM Configuration
Configure vLLM server parameters. Leave empty to use defaults.
Maximum context length the model can handle
Fraction of GPU memory to use (0.1 - 1.0)
GPT OSS 120B
GPT OSS 120B
Quick Start Runner
Inference Engine
Loading command...Commands are auto-generated based on your configuration settings.
Advanced configuration
Authentication
This model requires a Hugging Face access token. The token is inserted into the command and never stored.
vLLM Configuration
Configure vLLM server parameters. Leave empty to use defaults.
Maximum context length the model can handle
Fraction of GPU memory to use (0.1 - 1.0)
Alibaba Qwen3
Qwen3 4B
Qwen3 4B
Quick Start Runner
Inference Engine
Loading command...Commands are auto-generated based on your configuration settings.
Advanced configuration
Authentication
This model requires a Hugging Face access token. The token is inserted into the command and never stored.
vLLM Configuration
Configure vLLM server parameters. Leave empty to use defaults.
Maximum context length the model can handle
Fraction of GPU memory to use (0.1 - 1.0)
Qwen3 8B
Qwen3 8B
Quick Start Runner
Inference Engine
Loading command...Commands are auto-generated based on your configuration settings.
Advanced configuration
Authentication
This model requires a Hugging Face access token. The token is inserted into the command and never stored.
vLLM Configuration
Configure vLLM server parameters. Leave empty to use defaults.
Maximum context length the model can handle
Fraction of GPU memory to use (0.1 - 1.0)
Qwen3 30B-A3B (MoE)
Qwen3 30B-A3B (MoE)
Quick Start Runner
Inference Engine
Loading command...Commands are auto-generated based on your configuration settings.
Advanced configuration
Authentication
This model requires a Hugging Face access token. The token is inserted into the command and never stored.
vLLM Configuration
Configure vLLM server parameters. Leave empty to use defaults.
Maximum context length the model can handle
Fraction of GPU memory to use (0.1 - 1.0)
Qwen3 32B
Qwen3 32B
Quick Start Runner
Inference Engine
Loading command...Commands are auto-generated based on your configuration settings.
Advanced configuration
Authentication
This model requires a Hugging Face access token. The token is inserted into the command and never stored.
vLLM Configuration
Configure vLLM server parameters. Leave empty to use defaults.
Maximum context length the model can handle
Fraction of GPU memory to use (0.1 - 1.0)
Qwen3 VL 4B
Qwen3 VL 4B
Quick Start Runner
Inference Engine
Loading command...Commands are auto-generated based on your configuration settings.
Advanced configuration
Authentication
This model requires a Hugging Face access token. The token is inserted into the command and never stored.
vLLM Configuration
Configure vLLM server parameters. Leave empty to use defaults.
Maximum context length the model can handle
Fraction of GPU memory to use (0.1 - 1.0)
Qwen3 VL 8B
Qwen3 VL 8B
Quick Start Runner
Inference Engine
Loading command...Commands are auto-generated based on your configuration settings.
Advanced configuration
Authentication
This model requires a Hugging Face access token. The token is inserted into the command and never stored.
vLLM Configuration
Configure vLLM server parameters. Leave empty to use defaults.
Maximum context length the model can handle
Fraction of GPU memory to use (0.1 - 1.0)
Mistral AI Ministral 3
Ministral 3 3B Instruct
Ministral 3 3B Instruct
Quick Start Runner
Inference Engine
Loading command...Commands are auto-generated based on your configuration settings.
Advanced configuration
Authentication
This model requires a Hugging Face access token. The token is inserted into the command and never stored.
vLLM Configuration
Configure vLLM server parameters. Leave empty to use defaults.
Maximum context length the model can handle
Fraction of GPU memory to use (0.1 - 1.0)
Ministral 3 8B Instruct
Ministral 3 8B Instruct
Quick Start Runner
Inference Engine
Loading command...Commands are auto-generated based on your configuration settings.
Advanced configuration
Authentication
This model requires a Hugging Face access token. The token is inserted into the command and never stored.
vLLM Configuration
Configure vLLM server parameters. Leave empty to use defaults.
Maximum context length the model can handle
Fraction of GPU memory to use (0.1 - 1.0)
Ministral 3 14B Instruct
Ministral 3 14B Instruct
Quick Start Runner
Inference Engine
Loading command...Commands are auto-generated based on your configuration settings.
Advanced configuration
Authentication
This model requires a Hugging Face access token. The token is inserted into the command and never stored.
vLLM Configuration
Configure vLLM server parameters. Leave empty to use defaults.
Maximum context length the model can handle
Fraction of GPU memory to use (0.1 - 1.0)
Ministral 3 3B Reasoning
Ministral 3 3B Reasoning
Quick Start Runner
Inference Engine
Loading command...Commands are auto-generated based on your configuration settings.
Advanced configuration
Authentication
This model requires a Hugging Face access token. The token is inserted into the command and never stored.
vLLM Configuration
Configure vLLM server parameters. Leave empty to use defaults.
Maximum context length the model can handle
Fraction of GPU memory to use (0.1 - 1.0)
Ministral 3 8B Reasoning
Ministral 3 8B Reasoning
Quick Start Runner
Inference Engine
Loading command...Commands are auto-generated based on your configuration settings.
Advanced configuration
Authentication
This model requires a Hugging Face access token. The token is inserted into the command and never stored.
vLLM Configuration
Configure vLLM server parameters. Leave empty to use defaults.
Maximum context length the model can handle
Fraction of GPU memory to use (0.1 - 1.0)
Ministral 3 14B Reasoning
Ministral 3 14B Reasoning
Quick Start Runner
Inference Engine
Loading command...Commands are auto-generated based on your configuration settings.
Advanced configuration
Authentication
This model requires a Hugging Face access token. The token is inserted into the command and never stored.
vLLM Configuration
Configure vLLM server parameters. Leave empty to use defaults.
Maximum context length the model can handle
Fraction of GPU memory to use (0.1 - 1.0)
Meta Llama 3
Llama 3.2 3B
Llama 3.2 3B
Quick Start Runner
Inference Engine
Loading command...Commands are auto-generated based on your configuration settings.
Advanced configuration
Authentication
This model requires a Hugging Face access token. The token is inserted into the command and never stored.
vLLM Configuration
Configure vLLM server parameters. Leave empty to use defaults.
Maximum context length the model can handle
Fraction of GPU memory to use (0.1 - 1.0)
Llama 3.1 8B
Llama 3.1 8B
Quick Start Runner
Inference Engine
Loading command...Commands are auto-generated based on your configuration settings.
Advanced configuration
Authentication
This model requires a Hugging Face access token. The token is inserted into the command and never stored.
vLLM Configuration
Configure vLLM server parameters. Leave empty to use defaults.
Maximum context length the model can handle
Fraction of GPU memory to use (0.1 - 1.0)
Llama 3.1 70B
Llama 3.1 70B
Quick Start Runner
Inference Engine
Loading command...Commands are auto-generated based on your configuration settings.
Advanced configuration
Authentication
This model requires a Hugging Face access token. The token is inserted into the command and never stored.
vLLM Configuration
Configure vLLM server parameters. Leave empty to use defaults.
Maximum context length the model can handle
Fraction of GPU memory to use (0.1 - 1.0)
Google Gemma3
FunctionGemma
FunctionGemma
Quick Start Runner
Inference Engine
Loading command...Commands are auto-generated based on your configuration settings.
Advanced configuration
Authentication
This model requires a Hugging Face access token. The token is inserted into the command and never stored.
vLLM Configuration
Configure vLLM server parameters. Leave empty to use defaults.
Maximum context length the model can handle
Fraction of GPU memory to use (0.1 - 1.0)
Gemma 3 270M
Gemma 3 270M
Quick Start Runner
Inference Engine
Loading command...Commands are auto-generated based on your configuration settings.
Advanced configuration
Authentication
This model requires a Hugging Face access token. The token is inserted into the command and never stored.
vLLM Configuration
Configure vLLM server parameters. Leave empty to use defaults.
Maximum context length the model can handle
Fraction of GPU memory to use (0.1 - 1.0)
Gemma 3 1B
Gemma 3 1B
Quick Start Runner
Inference Engine
Loading command...Commands are auto-generated based on your configuration settings.
Advanced configuration
Authentication
This model requires a Hugging Face access token. The token is inserted into the command and never stored.
vLLM Configuration
Configure vLLM server parameters. Leave empty to use defaults.
Maximum context length the model can handle
Fraction of GPU memory to use (0.1 - 1.0)
Gemma 3 4B
Gemma 3 4B
Quick Start Runner
Inference Engine
Loading command...Commands are auto-generated based on your configuration settings.
Advanced configuration
Authentication
This model requires a Hugging Face access token. The token is inserted into the command and never stored.
vLLM Configuration
Configure vLLM server parameters. Leave empty to use defaults.
Maximum context length the model can handle
Fraction of GPU memory to use (0.1 - 1.0)
Gemma 3 12B
Gemma 3 12B
Quick Start Runner
Inference Engine
Loading command...Commands are auto-generated based on your configuration settings.
Advanced configuration
Authentication
This model requires a Hugging Face access token. The token is inserted into the command and never stored.
vLLM Configuration
Configure vLLM server parameters. Leave empty to use defaults.
Maximum context length the model can handle
Fraction of GPU memory to use (0.1 - 1.0)
Gemma 3 27B
Gemma 3 27B
Quick Start Runner
Inference Engine
Loading command...Commands are auto-generated based on your configuration settings.
Advanced configuration
Authentication
This model requires a Hugging Face access token. The token is inserted into the command and never stored.
vLLM Configuration
Configure vLLM server parameters. Leave empty to use defaults.
Maximum context length the model can handle
Fraction of GPU memory to use (0.1 - 1.0)
Black Forest Labs
FLUX 3 Action New
Alibaba Qwen3.8
Qwen3.8 Flash Next New
Qwen3.8 Flash Next
Quick Start Runner
Inference Engine
Loading command...Commands are auto-generated based on your configuration settings.
Advanced configuration
Authentication
This model requires a Hugging Face access token. The token is inserted into the command and never stored.
vLLM Configuration
Configure vLLM server parameters. Leave empty to use defaults.
Maximum context length the model can handle
Fraction of GPU memory to use (0.1 - 1.0)
Qwen3.8 27B New
Qwen3.8 27B
Quick Start Runner
Inference Engine
Loading command...Commands are auto-generated based on your configuration settings.
Advanced configuration
Authentication
This model requires a Hugging Face access token. The token is inserted into the command and never stored.
vLLM Configuration
Configure vLLM server parameters. Leave empty to use defaults.
Maximum context length the model can handle
Fraction of GPU memory to use (0.1 - 1.0)
MiniMax M2.7
MiniMax M2.7
MiniMax M2.7
Quick Start Runner
Inference Engine
Loading command...Commands are auto-generated based on your configuration settings.
Advanced configuration
Authentication
This model requires a Hugging Face access token. The token is inserted into the command and never stored.
vLLM Configuration
Configure vLLM server parameters. Leave empty to use defaults.
Maximum context length the model can handle
Fraction of GPU memory to use (0.1 - 1.0)
NVIDIA Cosmos
Cosmos3 Edge
Cosmos3 Edge
Quick Start Runner
Inference Engine
Loading command...Commands are auto-generated based on your configuration settings.
Advanced configuration
Authentication
This model requires a Hugging Face access token. The token is inserted into the command and never stored.
vLLM Configuration
Configure vLLM server parameters. Leave empty to use defaults.
Maximum context length the model can handle
Fraction of GPU memory to use (0.1 - 1.0)
Cosmos3 Nano
| | Modules | Inference engines | Actions | |||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|
| FLUX 3 Action New | Black Forest Labs | VLM | — | — | — | — | — | — | — | — | — | |
| Qwen3.8 Flash Next New | Alibaba Qwen3.8 | VLM | ✓ | — | — | — | — | ✓ | — | — | — | Qwen3.8 Flash NextQuick Start Runner Inference EngineCommands are auto-generated based on your configuration settings. Advanced configurationAuthenticationThis model requires a Hugging Face access token. The token is inserted into the command and never stored. vLLM ConfigurationConfigure vLLM server parameters. Leave empty to use defaults. Maximum context length the model can handle Fraction of GPU memory to use (0.1 - 1.0) |
| Qwen3.8 27B New | Alibaba Qwen3.8 | VLM | ✓ | ✓ | ✓ | — | — | — | — | ✓ | — | Qwen3.8 27BQuick Start Runner Inference EngineCommands are auto-generated based on your configuration settings. Advanced configurationAuthenticationThis model requires a Hugging Face access token. The token is inserted into the command and never stored. vLLM ConfigurationConfigure vLLM server parameters. Leave empty to use defaults. Maximum context length the model can handle Fraction of GPU memory to use (0.1 - 1.0) |
| Nemotron 3.5 Lightning | NVIDIA Nemotron | — | ✓ | ✓ | ✓ | — | — | ✓ | — | ✓ | ✓ | Nemotron 3.5 LightningQuick Start Runner Inference EngineCommands are auto-generated based on your configuration settings. Advanced configurationAuthenticationThis model requires a Hugging Face access token. The token is inserted into the command and never stored. vLLM ConfigurationConfigure vLLM server parameters. Leave empty to use defaults. Maximum context length the model can handle Fraction of GPU memory to use (0.1 - 1.0) |
| Muse Glimmer 30B | Meta Muse | VLM | ✓ | ✓ | ✓ | — | — | — | — | ✓ | — | Muse Glimmer 30BQuick Start Runner Inference EngineCommands are auto-generated based on your configuration settings. Advanced configurationAuthenticationThis model requires a Hugging Face access token. The token is inserted into the command and never stored. vLLM ConfigurationConfigure vLLM server parameters. Leave empty to use defaults. Maximum context length the model can handle Fraction of GPU memory to use (0.1 - 1.0) |
| Nemotron3 Nano 4B | NVIDIA Nemotron | — | ✓ | ✓ | ✓ | ✓ | ✓ | — | — | ✓ | — | Nemotron3 Nano 4BQuick Start Runner Inference EngineCommands are auto-generated based on your configuration settings. Advanced configurationAuthenticationThis model requires a Hugging Face access token. The token is inserted into the command and never stored. vLLM ConfigurationConfigure vLLM server parameters. Leave empty to use defaults. Maximum context length the model can handle Fraction of GPU memory to use (0.1 - 1.0) |
| FunctionGemma | Google Gemma3 | — | ✓ | ✓ | ✓ | ✓ | ✓ | — | — | ✓ | — | FunctionGemmaQuick Start Runner Inference EngineCommands are auto-generated based on your configuration settings. Advanced configurationAuthenticationThis model requires a Hugging Face access token. The token is inserted into the command and never stored. vLLM ConfigurationConfigure vLLM server parameters. Leave empty to use defaults. Maximum context length the model can handle Fraction of GPU memory to use (0.1 - 1.0) |
| Cosmos Reason 1 7B | NVIDIA Cosmos Reason | VLM | ✓ | ✓ | ✓ | ✓ | ✓ | ✓ | — | — | — | Cosmos Reason 1 7BQuick Start Runner Inference EngineCommands are auto-generated based on your configuration settings. Advanced configurationAuthenticationThis model requires a Hugging Face access token. The token is inserted into the command and never stored. vLLM ConfigurationConfigure vLLM server parameters. Leave empty to use defaults. Maximum context length the model can handle Fraction of GPU memory to use (0.1 - 1.0) |
| Gemma 3 270M | Google Gemma3 | — | ✓ | ✓ | ✓ | ✓ | ✓ | ✓ | ✓ | — | — | Gemma 3 270MQuick Start Runner Inference EngineCommands are auto-generated based on your configuration settings. Advanced configurationAuthenticationThis model requires a Hugging Face access token. The token is inserted into the command and never stored. vLLM ConfigurationConfigure vLLM server parameters. Leave empty to use defaults. Maximum context length the model can handle Fraction of GPU memory to use (0.1 - 1.0) |
| Gemma 4 E2B | Google Gemma4 | VLM | ✓ | ✓ | ✓ | ✓ | ✓ | ✓ | — | ✓ | — | Gemma 4 E2BQuick Start Runner Inference EngineCommands are auto-generated based on your configuration settings. Advanced configurationAuthenticationThis model requires a Hugging Face access token. The token is inserted into the command and never stored. vLLM ConfigurationConfigure vLLM server parameters. Leave empty to use defaults. Maximum context length the model can handle Fraction of GPU memory to use (0.1 - 1.0) |
| GPT OSS 20B | OpenAI GPT OSS | — | ✓ | ✓ | ✓ | — | — | ✓ | — | — | — | GPT OSS 20BQuick Start Runner Inference EngineCommands are auto-generated based on your configuration settings. Advanced configurationAuthenticationThis model requires a Hugging Face access token. The token is inserted into the command and never stored. vLLM ConfigurationConfigure vLLM server parameters. Leave empty to use defaults. Maximum context length the model can handle Fraction of GPU memory to use (0.1 - 1.0) |
| Llama 3.2 3B | Meta Llama 3 | — | ✓ | ✓ | ✓ | ✓ | ✓ | ✓ | ✓ | — | — | Llama 3.2 3BQuick Start Runner Inference EngineCommands are auto-generated based on your configuration settings. Advanced configurationAuthenticationThis model requires a Hugging Face access token. The token is inserted into the command and never stored. vLLM ConfigurationConfigure vLLM server parameters. Leave empty to use defaults. Maximum context length the model can handle Fraction of GPU memory to use (0.1 - 1.0) |
| MiniMax M2.7 | MiniMax M2.7 | — | ✓ | — | — | — | — | — | — | ✓ | — | MiniMax M2.7Quick Start Runner Inference EngineCommands are auto-generated based on your configuration settings. Advanced configurationAuthenticationThis model requires a Hugging Face access token. The token is inserted into the command and never stored. vLLM ConfigurationConfigure vLLM server parameters. Leave empty to use defaults. Maximum context length the model can handle Fraction of GPU memory to use (0.1 - 1.0) |
| Ministral 3 3B Instruct | Mistral AI Ministral 3 | VLM | ✓ | ✓ | ✓ | ✓ | ✓ | ✓ | ✓ | — | — | Ministral 3 3B InstructQuick Start Runner Inference EngineCommands are auto-generated based on your configuration settings. Advanced configurationAuthenticationThis model requires a Hugging Face access token. The token is inserted into the command and never stored. vLLM ConfigurationConfigure vLLM server parameters. Leave empty to use defaults. Maximum context length the model can handle Fraction of GPU memory to use (0.1 - 1.0) |
| Nemotron3 Nano 30B-A3B | NVIDIA Nemotron | — | ✓ | ✓ | ✓ | ✓ | ✓ | ✓ | ✓ | — | — | Nemotron3 Nano 30B-A3BQuick Start Runner Inference EngineCommands are auto-generated based on your configuration settings. Advanced configurationAuthenticationThis model requires a Hugging Face access token. The token is inserted into the command and never stored. vLLM ConfigurationConfigure vLLM server parameters. Leave empty to use defaults. Maximum context length the model can handle Fraction of GPU memory to use (0.1 - 1.0) |
| Qwen3 4B | Alibaba Qwen3 | — | ✓ | ✓ | ✓ | ✓ | ✓ | ✓ | — | — | — | Qwen3 4BQuick Start Runner Inference EngineCommands are auto-generated based on your configuration settings. Advanced configurationAuthenticationThis model requires a Hugging Face access token. The token is inserted into the command and never stored. vLLM ConfigurationConfigure vLLM server parameters. Leave empty to use defaults. Maximum context length the model can handle Fraction of GPU memory to use (0.1 - 1.0) |
| Qwen3.5 35B-A3B (MoE) | Alibaba Qwen3.5 | — | ✓ | ✓ | ✓ | — | — | ✓ | — | — | — | Qwen3.5 35B-A3B (MoE)Quick Start Runner Inference EngineCommands are auto-generated based on your configuration settings. Advanced configurationAuthenticationThis model requires a Hugging Face access token. The token is inserted into the command and never stored. vLLM ConfigurationConfigure vLLM server parameters. Leave empty to use defaults. Maximum context length the model can handle Fraction of GPU memory to use (0.1 - 1.0) |
| Qwen3.6 35B-A3B (MoE) | Alibaba Qwen3.6 | — | ✓ | ✓ | ✓ | — | — | ✓ | — | — | — | Qwen3.6 35B-A3B (MoE)Quick Start Runner Inference EngineCommands are auto-generated based on your configuration settings. Advanced configurationAuthenticationThis model requires a Hugging Face access token. The token is inserted into the command and never stored. vLLM ConfigurationConfigure vLLM server parameters. Leave empty to use defaults. Maximum context length the model can handle Fraction of GPU memory to use (0.1 - 1.0) |
| Gemma 3 1B | Google Gemma3 | — | ✓ | ✓ | ✓ | ✓ | ✓ | ✓ | ✓ | — | — | Gemma 3 1BQuick Start Runner Inference EngineCommands are auto-generated based on your configuration settings. Advanced configurationAuthenticationThis model requires a Hugging Face access token. The token is inserted into the command and never stored. vLLM ConfigurationConfigure vLLM server parameters. Leave empty to use defaults. Maximum context length the model can handle Fraction of GPU memory to use (0.1 - 1.0) |
| Gemma 4 12B | Google Gemma4 | VLM | ✓ | ✓ | ✓ | — | — | ✓ | — | ✓ | — | Gemma 4 12BQuick Start Runner Inference EngineCommands are auto-generated based on your configuration settings. Advanced configurationAuthenticationThis model requires a Hugging Face access token. The token is inserted into the command and never stored. vLLM ConfigurationConfigure vLLM server parameters. Leave empty to use defaults. Maximum context length the model can handle Fraction of GPU memory to use (0.1 - 1.0) |
| Gemma 4 E4B | Google Gemma4 | VLM | ✓ | ✓ | ✓ | ✓ | — | ✓ | — | ✓ | — | Gemma 4 E4BQuick Start Runner Inference EngineCommands are auto-generated based on your configuration settings. Advanced configurationAuthenticationThis model requires a Hugging Face access token. The token is inserted into the command and never stored. vLLM ConfigurationConfigure vLLM server parameters. Leave empty to use defaults. Maximum context length the model can handle Fraction of GPU memory to use (0.1 - 1.0) |
| GPT OSS 120B | OpenAI GPT OSS | — | ✓ | ✓ | — | — | — | ✓ | — | — | — | GPT OSS 120BQuick Start Runner Inference EngineCommands are auto-generated based on your configuration settings. Advanced configurationAuthenticationThis model requires a Hugging Face access token. The token is inserted into the command and never stored. vLLM ConfigurationConfigure vLLM server parameters. Leave empty to use defaults. Maximum context length the model can handle Fraction of GPU memory to use (0.1 - 1.0) |
| Llama 3.1 8B | Meta Llama 3 | — | ✓ | ✓ | ✓ | ✓ | ✓ | ✓ | ✓ | — | — | Llama 3.1 8BQuick Start Runner Inference EngineCommands are auto-generated based on your configuration settings. Advanced configurationAuthenticationThis model requires a Hugging Face access token. The token is inserted into the command and never stored. vLLM ConfigurationConfigure vLLM server parameters. Leave empty to use defaults. Maximum context length the model can handle Fraction of GPU memory to use (0.1 - 1.0) |
| Ministral 3 8B Instruct | Mistral AI Ministral 3 | VLM | ✓ | ✓ | ✓ | ✓ | ✓ | ✓ | ✓ | — | — | Ministral 3 8B InstructQuick Start Runner Inference EngineCommands are auto-generated based on your configuration settings. Advanced configurationAuthenticationThis model requires a Hugging Face access token. The token is inserted into the command and never stored. vLLM ConfigurationConfigure vLLM server parameters. Leave empty to use defaults. Maximum context length the model can handle Fraction of GPU memory to use (0.1 - 1.0) |
| Nemotron Nano 9B v2 | NVIDIA Nemotron | — | ✓ | ✓ | — | — | — | ✓ | — | — | — | Nemotron Nano 9B v2Quick Start Runner Inference EngineCommands are auto-generated based on your configuration settings. Advanced configurationAuthenticationThis model requires a Hugging Face access token. The token is inserted into the command and never stored. vLLM ConfigurationConfigure vLLM server parameters. Leave empty to use defaults. Maximum context length the model can handle Fraction of GPU memory to use (0.1 - 1.0) |
| Qwen3.5 27B | Alibaba Qwen3.5 | — | ✓ | ✓ | ✓ | — | — | ✓ | — | — | — | Qwen3.5 27BQuick Start Runner Inference EngineCommands are auto-generated based on your configuration settings. Advanced configurationAuthenticationThis model requires a Hugging Face access token. The token is inserted into the command and never stored. vLLM ConfigurationConfigure vLLM server parameters. Leave empty to use defaults. Maximum context length the model can handle Fraction of GPU memory to use (0.1 - 1.0) |
| Qwen3.6 27B | Alibaba Qwen3.6 | — | ✓ | ✓ | ✓ | — | — | ✓ | — | — | — | Qwen3.6 27BQuick Start Runner Inference EngineCommands are auto-generated based on your configuration settings. Advanced configurationAuthenticationThis model requires a Hugging Face access token. The token is inserted into the command and never stored. vLLM ConfigurationConfigure vLLM server parameters. Leave empty to use defaults. Maximum context length the model can handle Fraction of GPU memory to use (0.1 - 1.0) |
| Qwen3 8B | Alibaba Qwen3 | — | ✓ | ✓ | ✓ | ✓ | — | ✓ | — | — | — | Qwen3 8BQuick Start Runner Inference EngineCommands are auto-generated based on your configuration settings. Advanced configurationAuthenticationThis model requires a Hugging Face access token. The token is inserted into the command and never stored. vLLM ConfigurationConfigure vLLM server parameters. Leave empty to use defaults. Maximum context length the model can handle Fraction of GPU memory to use (0.1 - 1.0) |
| Cosmos Reason 2 2B | NVIDIA Cosmos Reason | VLM | ✓ | ✓ | ✓ | ✓ | ✓ | ✓ | — | ✓ | — | |
| Gemma 3 4B | Google Gemma3 | VLM | ✓ | ✓ | ✓ | ✓ | ✓ | ✓ | ✓ | — | — | Gemma 3 4BQuick Start Runner Inference EngineCommands are auto-generated based on your configuration settings. Advanced configurationAuthenticationThis model requires a Hugging Face access token. The token is inserted into the command and never stored. vLLM ConfigurationConfigure vLLM server parameters. Leave empty to use defaults. Maximum context length the model can handle Fraction of GPU memory to use (0.1 - 1.0) |
| Gemma 4 26B-A4B | Google Gemma4 | VLM | ✓ | ✓ | ✓ | — | — | ✓ | — | ✓ | — | Gemma 4 26B-A4BQuick Start Runner Inference EngineCommands are auto-generated based on your configuration settings. Advanced configurationAuthenticationThis model requires a Hugging Face access token. The token is inserted into the command and never stored. vLLM ConfigurationConfigure vLLM server parameters. Leave empty to use defaults. Maximum context length the model can handle Fraction of GPU memory to use (0.1 - 1.0) |
| Llama 3.1 70B | Meta Llama 3 | — | ✓ | ✓ | ✓ | ✓ | ✓ | ✓ | ✓ | — | — | Llama 3.1 70BQuick Start Runner Inference EngineCommands are auto-generated based on your configuration settings. Advanced configurationAuthenticationThis model requires a Hugging Face access token. The token is inserted into the command and never stored. vLLM ConfigurationConfigure vLLM server parameters. Leave empty to use defaults. Maximum context length the model can handle Fraction of GPU memory to use (0.1 - 1.0) |
| Ministral 3 14B Instruct | Mistral AI Ministral 3 | VLM | ✓ | ✓ | ✓ | ✓ | ✓ | ✓ | ✓ | — | — | Ministral 3 14B InstructQuick Start Runner Inference EngineCommands are auto-generated based on your configuration settings. Advanced configurationAuthenticationThis model requires a Hugging Face access token. The token is inserted into the command and never stored. vLLM ConfigurationConfigure vLLM server parameters. Leave empty to use defaults. Maximum context length the model can handle Fraction of GPU memory to use (0.1 - 1.0) |
| Nemotron 3 Super 120B-A12B | NVIDIA Nemotron | — | ✓ | — | — | — | — | ✓ | — | ✓ | — | Nemotron 3 Super 120B-A12BQuick Start Runner Inference EngineCommands are auto-generated based on your configuration settings. Advanced configurationAuthenticationThis model requires a Hugging Face access token. The token is inserted into the command and never stored. vLLM ConfigurationConfigure vLLM server parameters. Leave empty to use defaults. Maximum context length the model can handle Fraction of GPU memory to use (0.1 - 1.0) |
| Nemotron Nano 12B VL | NVIDIA Nemotron | VLM | ✓ | ✓ | — | — | — | ✓ | — | — | — | Nemotron Nano 12B VLQuick Start Runner Inference EngineCommands are auto-generated based on your configuration settings. Advanced configurationAuthenticationThis model requires a Hugging Face access token. The token is inserted into the command and never stored. vLLM ConfigurationConfigure vLLM server parameters. Leave empty to use defaults. Maximum context length the model can handle Fraction of GPU memory to use (0.1 - 1.0) |
| Qwen3 30B-A3B (MoE) | Alibaba Qwen3 | — | ✓ | ✓ | ✓ | — | — | ✓ | — | — | — | Qwen3 30B-A3B (MoE)Quick Start Runner Inference EngineCommands are auto-generated based on your configuration settings. Advanced configurationAuthenticationThis model requires a Hugging Face access token. The token is inserted into the command and never stored. vLLM ConfigurationConfigure vLLM server parameters. Leave empty to use defaults. Maximum context length the model can handle Fraction of GPU memory to use (0.1 - 1.0) |
| Qwen3.5 9B | Alibaba Qwen3.5 | VLM | ✓ | ✓ | ✓ | ✓ | — | ✓ | — | — | — | Qwen3.5 9BQuick Start Runner Inference EngineCommands are auto-generated based on your configuration settings. Advanced configurationAuthenticationThis model requires a Hugging Face access token. The token is inserted into the command and never stored. vLLM ConfigurationConfigure vLLM server parameters. Leave empty to use defaults. Maximum context length the model can handle Fraction of GPU memory to use (0.1 - 1.0) |
| Cosmos3 Edge | NVIDIA Cosmos | VLM | ✓ | ✓ | ✓ | — | — | ✓ | — | — | — | Cosmos3 EdgeQuick Start Runner Inference EngineCommands are auto-generated based on your configuration settings. Advanced configurationAuthenticationThis model requires a Hugging Face access token. The token is inserted into the command and never stored. vLLM ConfigurationConfigure vLLM server parameters. Leave empty to use defaults. Maximum context length the model can handle Fraction of GPU memory to use (0.1 - 1.0) |
| Gemma 3 12B | Google Gemma3 | VLM | ✓ | ✓ | ✓ | ✓ | ✓ | ✓ | ✓ | — | — | Gemma 3 12BQuick Start Runner Inference EngineCommands are auto-generated based on your configuration settings. Advanced configurationAuthenticationThis model requires a Hugging Face access token. The token is inserted into the command and never stored. vLLM ConfigurationConfigure vLLM server parameters. Leave empty to use defaults. Maximum context length the model can handle Fraction of GPU memory to use (0.1 - 1.0) |
| Cosmos Reason 2 8B | NVIDIA Cosmos Reason | VLM | ✓ | ✓ | ✓ | ✓ | ✓ | ✓ | — | ✓ | — | |
| Gemma 4 31B | Google Gemma4 | VLM | ✓ | ✓ | ✓ | — | — | ✓ | — | ✓ | — | Gemma 4 31BQuick Start Runner Inference EngineCommands are auto-generated based on your configuration settings. Advanced configurationAuthenticationThis model requires a Hugging Face access token. The token is inserted into the command and never stored. vLLM ConfigurationConfigure vLLM server parameters. Leave empty to use defaults. Maximum context length the model can handle Fraction of GPU memory to use (0.1 - 1.0) |
| Ministral 3 3B Reasoning | Mistral AI Ministral 3 | VLM | ✓ | ✓ | ✓ | ✓ | ✓ | ✓ | — | — | — | Ministral 3 3B ReasoningQuick Start Runner Inference EngineCommands are auto-generated based on your configuration settings. Advanced configurationAuthenticationThis model requires a Hugging Face access token. The token is inserted into the command and never stored. vLLM ConfigurationConfigure vLLM server parameters. Leave empty to use defaults. Maximum context length the model can handle Fraction of GPU memory to use (0.1 - 1.0) |
| Qwen3 32B | Alibaba Qwen3 | — | ✓ | ✓ | — | — | — | ✓ | — | — | — | Qwen3 32BQuick Start Runner Inference EngineCommands are auto-generated based on your configuration settings. Advanced configurationAuthenticationThis model requires a Hugging Face access token. The token is inserted into the command and never stored. vLLM ConfigurationConfigure vLLM server parameters. Leave empty to use defaults. Maximum context length the model can handle Fraction of GPU memory to use (0.1 - 1.0) |
| Nemotron 3 Nano Omni | NVIDIA Nemotron | VLM | ✓ | ✓ | ✓ | — | — | ✓ | ✓ | ✓ | — | Nemotron 3 Nano OmniQuick Start Runner Inference EngineCommands are auto-generated based on your configuration settings. Advanced configurationAuthenticationThis model requires a Hugging Face access token. The token is inserted into the command and never stored. vLLM ConfigurationConfigure vLLM server parameters. Leave empty to use defaults. Maximum context length the model can handle Fraction of GPU memory to use (0.1 - 1.0) |
| Qwen3.5 4B | Alibaba Qwen3.5 | VLM | ✓ | ✓ | ✓ | ✓ | ✓ | ✓ | — | — | — | Qwen3.5 4BQuick Start Runner Inference EngineCommands are auto-generated based on your configuration settings. Advanced configurationAuthenticationThis model requires a Hugging Face access token. The token is inserted into the command and never stored. vLLM ConfigurationConfigure vLLM server parameters. Leave empty to use defaults. Maximum context length the model can handle Fraction of GPU memory to use (0.1 - 1.0) |
| DiffusionGemma 26B-A4B | Google Gemma4 | — | ✓ | ✓ | ✓ | — | — | ✓ | — | — | — | DiffusionGemma 26B-A4BQuick Start Runner Inference EngineCommands are auto-generated based on your configuration settings. Advanced configurationAuthenticationThis model requires a Hugging Face access token. The token is inserted into the command and never stored. vLLM ConfigurationConfigure vLLM server parameters. Leave empty to use defaults. Maximum context length the model can handle Fraction of GPU memory to use (0.1 - 1.0) |
| Cosmos3 Nano | NVIDIA Cosmos | VLM | ✓ | ✓ | — | — | — | ✓ | — | — | — | |
| Gemma 3 27B | Google Gemma3 | VLM | ✓ | ✓ | ✓ | ✓ | ✓ | ✓ | ✓ | — | — | Gemma 3 27BQuick Start Runner Inference EngineCommands are auto-generated based on your configuration settings. Advanced configurationAuthenticationThis model requires a Hugging Face access token. The token is inserted into the command and never stored. vLLM ConfigurationConfigure vLLM server parameters. Leave empty to use defaults. Maximum context length the model can handle Fraction of GPU memory to use (0.1 - 1.0) |
| Ministral 3 8B Reasoning | Mistral AI Ministral 3 | VLM | ✓ | ✓ | ✓ | ✓ | — | ✓ | — | — | — | Ministral 3 8B ReasoningQuick Start Runner Inference EngineCommands are auto-generated based on your configuration settings. Advanced configurationAuthenticationThis model requires a Hugging Face access token. The token is inserted into the command and never stored. vLLM ConfigurationConfigure vLLM server parameters. Leave empty to use defaults. Maximum context length the model can handle Fraction of GPU memory to use (0.1 - 1.0) |
| Qwen3.5 0.8B | Alibaba Qwen3.5 | VLM | ✓ | ✓ | ✓ | ✓ | ✓ | ✓ | — | — | — | Qwen3.5 0.8BQuick Start Runner Inference EngineCommands are auto-generated based on your configuration settings. Advanced configurationAuthenticationThis model requires a Hugging Face access token. The token is inserted into the command and never stored. vLLM ConfigurationConfigure vLLM server parameters. Leave empty to use defaults. Maximum context length the model can handle Fraction of GPU memory to use (0.1 - 1.0) |
| Qwen3 VL 4B | Alibaba Qwen3 | VLM | ✓ | ✓ | ✓ | ✓ | ✓ | ✓ | — | — | — | Qwen3 VL 4BQuick Start Runner Inference EngineCommands are auto-generated based on your configuration settings. Advanced configurationAuthenticationThis model requires a Hugging Face access token. The token is inserted into the command and never stored. vLLM ConfigurationConfigure vLLM server parameters. Leave empty to use defaults. Maximum context length the model can handle Fraction of GPU memory to use (0.1 - 1.0) |
| Ministral 3 14B Reasoning | Mistral AI Ministral 3 | VLM | ✓ | ✓ | ✓ | — | — | ✓ | — | — | — | Ministral 3 14B ReasoningQuick Start Runner Inference EngineCommands are auto-generated based on your configuration settings. Advanced configurationAuthenticationThis model requires a Hugging Face access token. The token is inserted into the command and never stored. vLLM ConfigurationConfigure vLLM server parameters. Leave empty to use defaults. Maximum context length the model can handle Fraction of GPU memory to use (0.1 - 1.0) |
| Qwen3 VL 8B | Alibaba Qwen3 | VLM | ✓ | ✓ | ✓ | ✓ | — | ✓ | — | — | — | Qwen3 VL 8BQuick Start Runner Inference EngineCommands are auto-generated based on your configuration settings. Advanced configurationAuthenticationThis model requires a Hugging Face access token. The token is inserted into the command and never stored. vLLM ConfigurationConfigure vLLM server parameters. Leave empty to use defaults. Maximum context length the model can handle Fraction of GPU memory to use (0.1 - 1.0) |
🔍
No models found
Try a different model name filter or clear column checkboxes in the table header
Performance Comparison
Benchmarks on Jetson across supported inference runtimes
Platform
Concurrency
No data for this platform
* ISL/OSL for all benchmarks: 2048/128
* Unless otherwise specified, all models utilize W4A16 quantization for Orin and NVFP4 for Thor.
* NVFP4 and MXFP4 require Blackwell FP4 tensor cores and are not available on Orin (Ampere).
* Edge-LLM benchmark values are generation throughput from the official TensorRT Edge-LLM performance benchmarks on Jetson AGX Thor Developer Kit.
* Cosmos3 Edge (Reasoner) (BF16): streaming chat harness, ISL 1705 / OSL 128, per-request prompt-cache salting, median of 3 - not the aiperf ISL-2048 protocol used for other rows.