Skip to main content
Language modelOpen weights

Holo4-35B-A3B

H Company

Released
September 28, 2026
Data date
October 3, 2026

Holo4-35B-A3B applies H Company’s software-interaction training to Qwen3.6-35B-A3B. Unlike the dense 27B variant, it uses a mixture-of-experts architecture with 35 billion total parameters and 3 billion active per token.

The model combines visual screen interaction with code and API tools. Its weights use Apache 2.0, and its context window holds 262,144 tokens.

Specifications and access

SpecificationValue and source
Model ID
Hcompany/Holo4-35B-A3BSource
Model class
Vision-language model for computer useSource
Architecture
Qwen3.6 mixture of expertsSource
Parameters
35 billion total, 3 billion activeSource
Context window
262,144 tokensSource
Input
Screenshots, text, and tool resultsSource
License
Apache 2.0Source