What happened
Holo4-35B-A3B is a 35-billion-parameter vision-language model built on Qwen3.6-35B-A3B. Its weights are Apache 2.0. Holotron4-30B-A3B is listed at about 33 billion parameters, uses NVIDIA Nemotron 3 Nano Omni and is governed by the NVIDIA Open Model Agreement. Both accept images and text for computer-use actions through H Company's agent harness.
The cards report a 262,144-token context for Holotron4 and provide downloadable BF16 weights plus quantized variants. H Company's benchmark scores and cost figures are developer-reported and do not establish production reliability. Neither card lists deployment by a hosted inference provider.
Why it matters
The two checkpoints widen architectural and licensing choices beyond the previously covered Holo4-27B. Downloadable weights are not the same as a safe autonomous system: tool permissions, prompt injection and action verification remain deployment responsibilities.
What to watch
Watch for independent computer-use evaluations, exact hardware requirements, commercial interpretation of the NVIDIA agreement and hosted availability.
Sources & verification
Source links are not a count of independent confirmations.
01Report a correction
Help us get it right
Include the disputed statement and, if possible, a supporting source URL. Please do not send personal or sensitive information.
