Holo4: powering generalist computer-use agents
AI Summary
Holo4 is a new series of agentic models available in 27B dense and 35B-A3B Mixture of Experts sizes, alongside Holotron4 Nano. Built on a Qwen base and trained using supervised and reinforcement learning on environments generated by an Agentic Task Factory, Holo4 interacts through GUIs, code sandboxes, MCP, and business APIs. It achieves competitive performance on long workflows like OSWorld 2.0 at a fraction of the cost of frontier models.
Key Takeaways
- Available in 27B dense, 35B-A3B MoE, and Holotron4 Nano sizes on the H Models API and Hugging Face.
- Trained on roughly 10,000 tasks built by the Agentic Task Factory.
- Scores 61.7% on OSWorld 2.0 with significantly fewer parameters than closed alternatives.
Why it matters
Holo4 provides engineers with open-weight, cost-effective generalist agentic models that seamlessly operate across multiple interfaces—including GUIs, code, and APIs—without requiring different models for each platform.
Conversation 0
Be the first to share your perspective.