Nemotron 3 Ultra
NVIDIA🇺🇸 USMajor lab· Jun 4, 2026
ReasoningTextopen weightsAPI✎ Editorial
NVIDIA's powerful open-weights MoE model: hybrid LatentMoE + MTP layers, 1M context, 550B/55B active. Free on OpenRouter. Focus: complex multi-agent workflows, code, math, science.
Editorial
- Context
- 1M tokens
- Params
- 550B MoE (55B active)
- License
- NVIDIA Open Model License
- Modalities
- text, code → text, code
3 variants
Sources
HuggingFaceOpenRouterBlogDeutsch
NVIDIAs mächtiges open-weights MoE-Modell: Hybrid LatentMoE + MTP-Layer, 1M Kontext, 550B/55B aktiv. Auf OpenRouter kostenlos nutzbar. Fokus: komplexe Multi-Agenten-Workflows, Code, Mathe, Wissenschaft.
Benchmarks
| Benchmark | Domain | Value | Evidence |
|---|---|---|---|
| LongBench v2vendor number | long-context | 61.9 % correct | 90 |
| LiveCodeBenchvendor number | coding | 89 % pass@1 | 80 |
| SWE-bench Verifiedvendor number | coding | 70.7 % solved | 71 |
| RULERvendor number | long-context | 94.7 % correct | 68 |
| MMLUvendor number | knowledge | 86.8 % correct | 43 |
Vendor-reported numbers are marked and count with factor 0.7 in rankings. The evidence score says how much a benchmark still tells you today.
Provenance
- Detected
- Jun 4, 2026
- First source
- kuratiert
- Description
- Editorial
- Editorially reviewed
- yes