Updated

Holo4

Hcompany released Holo4, a family of agentic models in two sizes: a 27B dense model and a 35B MoE (35B-A3B) model, plus an updated Holotron4 Nano. They work across GUIs, code, MCP, and APIs, and are available through the H Models API with open weights on Hugging Face.

By No Code MBA · Facts checked

Provider
Hcompany
Announcement date
Status at last check
Generally available
Model family
Holo4
Preceding model
Holo3.1
Capabilities
Text, Image, Coding

What changed

Holo4 comes in two sizes, Holo4-27B (dense, fine-tuned from Qwen3.8-27B) and Holo4-35B-A3B (mixture of experts), plus an updated Holotron4 Nano built on NVIDIA Nemotron 3 Nano Omni. The models take screenshots and text and act across GUIs, code sandboxes, MCP and APIs, with a 262,144-token context on the 27B. Hcompany reports Holo4-27B at 61.7% on OSWorld 2.0 versus 30.9% for the 35B-A3B, and says it trained with 127B tokens of supervised data plus reinforcement learning.

Access and availability

Both sizes are available through the H Models API (quickstart at hub.hcompany.ai). Open weights are on Hugging Face (Hcompany/Holo4-27B, Hcompany/Holo4-35B-A3B, Hcompany/Holotron4-30B-A3B), with Holo4-27B under the non-commercial CC BY-NC 4.0 license.

Official access documentation

What you can build

Evaluate Holo4 through the H Models API on a multi-step task that mixes screen interaction, code execution, and API tool calls, since it is built to switch between interfaces within one workflow.

Limitations

Hcompany says Holo4 trails the strongest closed models on long workflows, and its own comparisons mix harnesses and task subsets. The 27B weights are non-commercial, and AutomationBench private-set results were pending.

Sources and release history