Model licenses & attribution

Last updated July 10, 2026.

VxNewton is an independent application that runs Google Gemma 4 model components locally through Apple's MLX framework. The VxNewton application and the model components have separate licenses.

Gemma 4 license

The Gemma 4 target and MTP assistant components distributed for VxNewton are provided under the Apache License 2.0. VxNewton's terms do not replace or restrict the rights granted for those components by Apache 2.0.

Every model package includes the complete Apache 2.0 license and a provenance and modification notice. The installed app also makes its full third-party notices available from VxNewton > Acknowledgements.

Main model

VxNewton uses a pinned 4-bit MLX conversion of Google's instruction-tuned Gemma 4 E4B model:

Original model: google/gemma-4-E4B-it
MLX conversion: mlx-community/gemma-4-e4b-it-4bit, pinned revision
Conversion tool recorded by the publisher: mlx-vlm 0.4.3

MTP assistant

VxNewton's speculative-generation assistant uses the official Google Gemma 4 E4B MTP BF16 files. The packaged runtime files are byte-for-byte identical to Google's pinned official revision:

Official model: google/gemma-4-E4B-it-assistant, pinned revision
Runtime: loaded locally through MLX for speculative generation

Attribution and independence

Gemma is provided by Google. VxNewton is not affiliated with, sponsored by, or endorsed by Google. References to Google, Gemma, Hugging Face, MLX, and MLX Community identify component origins and do not imply endorsement.

The model packages and in-app acknowledgements are the authoritative license copies distributed with VxNewton. This page is a readable summary of that packaged attribution.