2026-09-28 — views
Hcompany Releases Holo4 Agentic Models for Generalist Computer Use
Read this because Holo4 offers a unified model for multi-interface automation, competing with frontier models at lower cost while open-sourcing training trajectories.
Hcompany introduces Holo4, a new series of agentic models supporting GUI, code, MCP, and API interfaces for generalist computer-use tasks.
Holo4: A Generalist Approach to Computer Use
Hcompany has released Holo4, a new series of agentic models designed to power generalist computer-use agents. The series includes two primary sizes: a 27B dense model and a 35B-A3B Mixture of Experts variant. Additionally, an updated version of their previous model, Holotron 3, is available as Holotron4 Nano. All models are accessible via the H Models API, with weights and full collections available in FP16, FP8, and GGUF formats on Hugging Face.
Multi-Interface Interaction
Unlike many agentic models that are trained for a single interface, Holo4 is designed to interact with software through any available method. It can click and type on graphical user interfaces (GUIs), write and execute code, or call MCP and API tools. The model adapts to the task at hand, running on desktops, web browsers, Android devices, code sandboxes, and against business APIs. Users do not need to select different models for different platforms; the same model handles the variety of interfaces required in real-world business workflows.
Performance and Cost Efficiency
Holo4 models are built upon Qwen bases and show significant improvements over their predecessors. On the OSWorld 2.0 benchmark, Holo4 27B scores 61.7%, trailing only the strongest closed models like Opus 5.5, which scores 81.8%. The 35B-A3B variant reaches 30.9%. Hcompany emphasizes that these results are achieved with orders of magnitude fewer parameters and at a much lower cost per task compared to frontier models. On the AutomationBench for API use, Holo4 also competes with leading closed models while maintaining lower operational costs.
Training and Capabilities
The models were trained through supervised and reinforcement learning on a large set of environments and tasks generated by Hcompany’s Agentic Task Factory. This internal pipeline creates interactive environments and verifiable tasks from documentation, such as screenshots of websites or open-source software. Holo4 demonstrates proficiency in professional software tasks, including complex 3D modeling in FreeCAD and game design in Godot. For instance, in a Pac-Man game design task, Holo4 27B required 68 calls and 2.4 million tokens, whereas the base Qwen3.8 27B model required 197 calls and 11.4 million tokens.
Holotron4 Nano and Open Data
Holo4 builds on a post-training stack that adapts to new foundation models. As part of the NVIDIA Nemotron Coalition, Hcompany applied this stack to the Nemotron 3 Nano Omni model to create Holotron4 Nano, which significantly improves over the base model on GUI workflows and environments exposing MCP, APIs, or coding sandboxes. Hcompany is also open-sourcing every trajectory behind their benchmark scores, allowing users to replay steps or download datasets from Hugging Face.
Practitioner note
Developers should evaluate Holo4 if they require a single model to handle diverse automation tasks across GUIs, APIs, and code without switching infrastructure. The availability of open-sourced trajectories allows for deep inspection of agent behavior and potential fine-tuning on specific business workflows.