Release of open-source agentic models Holo4 for GUI, code, and API automation
The new family of open-source agentic models Holo4 (27B dense, 35B-A3B MoE) handles GUI control, writing and running code, MCP, and API all in one model. On OSWorld 2.0, Holo4-27B achieves 61.7% compared to 81.8% for the Opus 5.5 model, but at significantly lower cost.
A new family of agentic models called Holo4 has been released, available in two sizes: 27B (dense) and 35B-A3B (Mixture of Experts). The models are accessible via the H Models API and as open source in FP16, FP8, and GGUF formats on Hugging Face. At the same time, an updated version of the smaller Holotron 3 model, called Holotron4 Nano, was released.
According to the manufacturer, Holo4 controls software through any available interface – graphical user interface (GUI), writing and running its own code, calling tools via MCP, and direct API – and the same model can be deployed without modification on desktop, web, Android, in a code sandbox, and against enterprise APIs. The model was trained using a combination of supervised learning and reinforcement learning on a large set of environments and tasks generated by the internal Agentic Task Factory system, using Qwen series models as its base (Qwen3.8 27B and Qwen3.6 35B-A3B).
On the OSWorld 2.0 benchmark, which tests long computer-use workflows, Holo4-27B achieved a score of 61.7% and Holo4-35B-A3B achieved 30.9%, while the competing model Opus 5.5 achieved 81.8%. According to the manufacturer, Holo4 achieves this performance with an order of magnitude fewer parameters and at a much lower cost per task than closed frontier models; the manufacturer also published complete trajectories from the benchmark runs for verification of the results.
The source article further describes comparative tests against the base model Qwen3.8 27B on specific tasks (modeling in FreeCAD, playing in Godot), but the full text of the rest of the article is not available.
Why it matters
For developers of agentic applications, this represents a cheaper and open alternative to closed frontier models – lower accuracy on long tasks, but lower operating costs and the ability to deploy locally thanks to open-source versions in FP16, FP8, and GGUF formats.
Two audiences, two different impacts
What this means
For individuals
Developers and individuals experimenting with agentic models gain a freely available alternative that can be run locally or via API, without having to pay for closed frontier models.
For a business
Companies building GUI, code, and API automation can deploy an open-source model with significantly lower per-task costs than closed frontier models, at the cost of lower accuracy on long agentic tasks.
ProductivityCheck the original
Event sources
clearly official source · 1 publisher, 0 independent. We count feeds from the same owner only once.