Gnani-Evon-v3.3-30B-A3B is a long-context model built on a hybrid Mamba-Transformer mixture-of-experts design, according to an arXiv preprint. The work is described as a sovereign, Indic-first reasoning model, and it starts from NVIDIA's Nemotron-3-Nano-30B-A3B base.
The model is intended for reasoning, instruction following, and tool use in English and eight Indian languages. The abstract does not specify which eight languages, nor does it include evaluation results or comparisons with other models.
Because this is a single-source announcement, there are no independent results or differing claims to weigh. The preprint is available on arXiv under identifier 2610.15371.