H company launches holo4: AI models that interact with computers across multiple platforms

H Company has launched Holo4, a new family of generalist computer-use AI models capable of clicking, typing, coding, and calling tools across desktop, web, Android, and APIs. The models come in two versions: Holo4 27B, a dense model, and Holo4 35B-A3B, a Mixture of Experts model with 3B active parameters. Both support a 256K context window via the H Models API. Holo4 is a vision-language model fine-tuned from Qwen3.8-27B and Qwen3.6-35B-A3B, and pairs with H’s open hai-agents harness, which processes screenshots and tool results to execute tasks. The models are designed to address gaps in existing AI agents, which often fail without screen access or API integration. Holo4 operates seamlessly across multiple platforms, including desktop, web, Android, and business APIs. Performance benchmarks show Holo4 27B scoring 85.2% on OSWorld at $0.08 per task, outperforming its base model Qwen3.8-27B. However, long workflows reveal challenges, with Holo4 27B scoring 61.7% on OSWorld 2.0 at $1.22 per task. H Company also released Holotron4 Nano, built on NVIDIA’s Nemotron 3 Nano Omni, which significantly improved OSWorld performance.