Models · Compare
Nemotron 3 Nano vs Gemini 3 Flash
Rows highlighted in warm gray are where the models differ. Numbers carry their as-of date and primary source.
Specs
| Field | A: Nemotron 3 Nano | B: Gemini 3 Flash |
|---|---|---|
| Released | 2025-12-15 | 2025-12-17 |
| Developer | NVIDIA | Google DeepMind |
| Openness | Open weights | Proprietary |
| License | NVIDIA Open Model License | Proprietary |
| OSI-approved | no | no |
| Data released | no | no |
| Training code | no | no |
| Architecture | hybrid-mamba-transformer-moe | unknown |
| Total params | 30B | — |
| Active params | — | — |
| Experts | — | — |
| Context window | 1.0M | 1.0M |
| Attention | hybrid-mamba2-gqa | unknown |
| Position enc. | unknown | unknown |
| Pretraining tokens | — | — |
| Post-training | sft, rlhf | rlhf |
| Training hardware | — | — |
| $/M input | — | $0.50 |
| $/M output | — | $3.00 |
| Output tok/sec | — | 185.9 |
Benchmarks
Missing scores render as not reported; never inferred. Bold highlights the leader per benchmark.
General reasoning
| MMLU-Pro | — | 88.2 2026-05-21 |
| GPQA-Diamond | — | 90.4 2025-12-17 |
Code
| SWE-Bench Verified | — | 78.0 2025-12-17 |
| LiveCodeBench | — | 79.7 2026-05-21 |
Math
| AIME 2025 | — | 55.7 2026-05-21 |
Context · A
First Nemotron 3 family member, a 31.6B-total / 3.2B-active hybrid Mamba-Transformer MoE. 52 layers (23 MoE, 23 Mamba-2, 6 GQA), 128 routed plus 1 shared expert with 6 activated per token. NVIDIA reports 3.3x Qwen3-30B-A3B throughput on a single H200 at 8K input / 16K output, with 1M-context support. Super and Ultra variants planned for H1 2026.
Context · B
Speed tier of the Gemini 3 family, released December 17 2025. Google reported 3x the speed of Gemini 2.5 Pro while outperforming it on coding and reasoning, with 30 percent fewer tokens on average for everyday tasks. Became the default in the Gemini app and rolled into Search AI Mode at launch.