Skip to main content
Language modelOpen weights

Holotron4-30B-A3B

H Company

Released
September 28, 2026
Data date
October 3, 2026

Holotron4-30B-A3B applies H Company’s agent training to NVIDIA’s Nemotron 3 Nano Omni. It therefore uses a different foundation from the Qwen-based Holo4 models. The post-training targets screen interaction, coding sandboxes, and environments with API or MCP tools.

The weights use the NVIDIA Open Model Agreement. The checkpoint’s name says 30B-A3B, while Hugging Face’s summary reports 33 billion parameters. Its model card specifies up to 262,144 context tokens.

Specifications and access

SpecificationValue and source
Model ID
Hcompany/Holotron4-30B-A3BSource
Model class
Vision-language model for computer useSource
Architecture
NemotronH Nano OmniSource
Parameters
The model name says 30B-A3B, while Hugging Face’s weights summary reports 33B parameters. A common counting basis for these figures is not documented.Source
Context window
262,144 tokensSource
Input
Screenshots, text, and tool resultsSource
License
NVIDIA Open Model AgreementSource