H Company just released Holotron-12B. Developed with NVIDIA, it's a high-throughput, open-source, multimodal model engineered specifically for the age of computer-use agents. (Performance on par with Holo2/Qwen but with 2x higher throughput)

H Company has launched Holotron-12B, a significant new open-source model developed in collaboration with NVIDIA. Engineered specifically for the emerging era of computer-use agents—AI systems that can interact with and control software interfaces—the 12-billion-parameter model is multimodal, meaning it can process both images and text. Its standout feature is a claimed 2x higher throughput compared to similar-performing models like Holo2 and Qwen, enabling faster data processing and response generation for agentic workflows.

This performance boost is critical for practical AI agent deployment, where speed directly impacts usability and cost-efficiency. By being open-source and available on Hugging Face, H Company lowers the barrier to entry for developers and researchers looking to experiment with or build advanced automation agents. The model's architecture, detailed in a technical report, is optimized for the high-volume, sequential tasks typical of agents operating in digital environments, from automating data entry to conducting complex multi-step research.