Holo4 clicks, codes, and calls APIs — one model

H Company's new agentic series runs across desktop, web, and Android without swapping models, and open-sources every benchmark run.

By The ai-compute Desk

[1][2]

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. [1] Holo4 sidesteps that constraint by running on desktops, on the web, on Android, in a code sandbox, and against business APIs — the same model, called the same way, in each environment. [2]

On OSWorld 2.0, Holo4 27B scores 61.7% against 81.8% for Opus 5.5, with orders of magnitude fewer parameters and at a much lower cost. [1][2] The headline number from H Company's own benchmarking is sharper still: Holo4 27B scores 85.2% on OSWorld at $0.08 per task. [2] Every trajectory behind the public benchmark scores is open-sourced and replayable at trajectories.hcompany.ai. [1]

[1][2]

The push toward interface-agnostic agents marks a quiet but meaningful shift in how agentic tooling is being designed — away from purpose-built wrappers and toward models that treat the choice of interface as just another decision to make mid-task.

[1][2]

Real work is not siloed that way, and a single business task can require combining these different approaches.H Company