---
title: "Holo4: powering generalist computer-use agents"
description: "Holo4 is a new series of agentic models designed to interact with software through any available interface: GUIs, code, MCP, and APIs. It comes in two sizes, a 27B dense model and a 35B-A3B Mixture of Experts model, both available on the H Models API. The models were trained through supervised and reinforcement learning on environments and tasks generated by the Agentic Task Factory, which produces verifiable tasks from documentation alone. On the OSWorld 2.0 benchmark, Holo4 27B scores 61.7% against 81.8% for Opus 5.5, achieving competitive performance at a fraction of the cost. The company also released Holotron4 Nano, an updated version of Holotron 3 built on the Nemotron 3 Nano Omni model using the same post-training stack."
---

# Holo4: powering generalist computer-use agents

[Hugging Face](https://yomu.fyi/company/hugging-face) · Tony Wu, Maxime Theillard, Frederic Renard, Vincent Coyette, Emrick Sinitambirivoutin, Avshalom Manevich, Antonio Loison, Antoine Bonnet, Maxime Langevin, Aleix Cambray (H-AI), Léonard Benedetti, Tony Wu, Mats L. Richter, Michael Eickenberg, Sławek Mucha, Matthias Brunel, Daniel Beechey · Sep 28, 2026

**Type:** Announcement

## Summary

Holo4 is a new series of agentic models designed to interact with software through any available interface: GUIs, code, MCP, and APIs. It comes in two sizes, a 27B dense model and a 35B-A3B Mixture of Experts model, both available on the H Models API. The models were trained through supervised and reinforcement learning on environments and tasks generated by the Agentic Task Factory, which produces verifiable tasks from documentation alone. On the OSWorld 2.0 benchmark, Holo4 27B scores 61.7% against 81.8% for Opus 5.5, achieving competitive performance at a fraction of the cost. The company also released Holotron4 Nano, an updated version of Holotron 3 built on the Nemotron 3 Nano Omni model using the same post-training stack.

## Context

Most agentic models are trained for one interface only: GUI-focused models are blind without a screen, while models that prefer tool calling are stuck in front of an application that has no API. Real business tasks often require combining GUI, code, MCP, and API interactions, which Holo4 is designed to handle with a single model.

## Approach / What changed

Holo4 was trained through supervised and reinforcement learning on a large set of environments and tasks, including those generated by the Agentic Task Factory, which builds interactive environments and verifiable tasks from documentation alone. The training harness was rebuilt using feedback from agentic performance on OSWorld 2.0, giving the agent reliable memory for hundreds of steps and a shell on the desktop machine.

## Takeaways

- Holo4 is a single model that works across desktops, the web, Android, code sandboxes, and business APIs without needing a different model per platform.
- On OSWorld 2.0, Holo4 27B scores 61.7% against 81.8% for Opus 5.5, achieving competitive performance with orders of magnitude fewer parameters and lower cost.
- The Agentic Task Factory automatically builds interactive environments and verifiable tasks from documentation alone, producing about 10,000 tasks across web apps, MCP servers, and desktop environments.

## FAQ

### What interfaces does Holo4 support?

Holo4 interacts with software through any available interface: GUIs, code, MCP, and APIs. It clicks and types on a screen, writes and runs its own code, and calls MCP or API tools, using whichever fits the task.

### What sizes and formats are available?

Holo4 comes in two sizes: a 27B dense model and a 35B-A3B Mixture of Experts model. Both are available on the H Models API, and weights are on Hugging Face in BF16, FP8, NVFP4, and 4-bit GGUF.

### How does Holo4 compare to frontier models on benchmarks?

On OSWorld 2.0, Holo4 27B scores 61.7% against 81.8% for Opus 5.5, and Holo4 35B-A3B reaches 30.9%. It does so with orders of magnitude fewer parameters and at a much lower cost per task.

**Tags:** [AI Agents](https://yomu.fyi/topic/ai-agents), [API](https://yomu.fyi/topic/api), [GUI](https://yomu.fyi/topic/gui), [MCP](https://yomu.fyi/topic/mcp), [Mixture of Experts](https://yomu.fyi/topic/mixture-of-experts)

- Source: [Hugging Face](https://huggingface.co/blog/Hcompany/holo4)
- Source URL: https://huggingface.co/blog/Hcompany/holo4
- Ingested by Yomu: 2026-09-28T12:00:10.119Z

[Read original post](https://huggingface.co/blog/Hcompany/holo4)