Side-by-side comparison of Gemma 4 26B A4B Instruct (Google DeepMind · United States) and Nemotron-3 Nano Omni 30B-A3B Reasoning (NVIDIA · USA) for self-hosted deployment of the open-weight model. Gemma 4 26B A4B Instruct is rated conditional; Nemotron-3 Nano Omni 30B-A3B Reasoning is conditional. They part ways on licence: Gemma 4 26B A4B Instruct is "Apache 2.0", Nemotron-3 Nano Omni 30B-A3B Reasoning is "NVIDIA Open Model".
| Field | ||
|---|---|---|
| Summary | ||
| Verdict | Conditional Based on published licence terms, Gemma 4 26B A4B ships under pure Apache 2.0 with no prohibited-use carve-outs — a departure from prior Gemma generations. The sparse-MoE architecture (25.2B total / 3.8B active) puts it in an ambiguous zone for EU AI Act GPAI systemic-risk classification, and US origin plus image-input support add transparency obligations that deployers should document. | Conditional Per the NVIDIA Open Model Agreement, Nemotron-3 Nano Omni is commercially usable with a NOTICE-file attribution requirement and U.S. export-compliance obligations. Multimodal MoE (31B total / 3B active) accepting video, audio, image and text input, with reasoning-style chain-of-thought output. Training data is unusually well-documented (1,395 datasets, modality breakdown, CSAM scanning) — useful for AI Act Article 53 mapping. Vendor jurisdiction remains the US. |
| Last reviewed | 2026-04-17 | 2026-05-03 |
| Open-weight | ||
| Licence | Apache 2.0 | NVIDIA Open Model |
| Commercial use | Unrestricted | Permitted (with attribution) |
| Training data | Domain-level summary | Disclosed |
| Origin | United States | USA |
| Performance & pricing? | ||
| Quality index | 27/100 | — |
| Speed | — | — |
| Blended price | — | — |
| Context window | — | — |
| Evidence | ||
| Sources | ||
No overlapping sources between the two entries.