Nemotron 3 Ultra
Released
2026-06-04
Last refreshed
2026-10-17
Status
Researched 28d ago
Open weightsCommercial use: permittedCodingRAGAgentsLong contextClassificationJSON / Tool useHighlight
NVIDIA AI releases · 27 in the last 12 months · this family litChangelog →
Specifications
- Family
- Nemotron 3
- Released
- 2026-06-04
- Context
- 1M
- Parameters
- 550B
- Architecture
- Mixture of Experts
- Specialization
- general
- Openness
- Open weights
- License
- OpenMDW 1.1Permitted
- Weights
- Available
- Code
- Unknown
Created by
Pricing
Output / 1M
$1.70
Input / 1M
$0.900
Cheapest of 8 routes · DigitalOcean GenAI · cache read $0.180
Providers(8)
View 8 provider routesLinks
Created by
Pricing
Output / 1M
$1.70
Input / 1M
$0.900
Cheapest of 8 routes · DigitalOcean GenAI · cache read $0.180
Providers(8)
View 8 provider routesLinks
About
NVIDIA's open frontier-reasoning model (550B total / 55B active MoE, hybrid Transformer-Mamba). 300+ tokens/second. 1M-token context. Announced at Computex 2026. Pricing: ~$0.60/$2.60 per 1M tokens (provider median); free tier on some providers.
Providers(8)
Compare all 8Compare API pricing across 8 tracked provider routes for input and output tokens, batch, and cached reads when available.
| Provider | Input / 1M | Output / 1M | Cache | Route |
|---|---|---|---|---|
| DigitalOcean GenAI | $0.900 | $1.70 | read $0.180 | Serverless |
| Weights & Biases | $0.500 | $2.15 | read $0.100 | Serverless |
| DeepInfra | $0.500 | $2.20 | read $0.100 / 5m $0.625 / 1h $1.00 | Serverless |
| Baseten API | $0.600 | $2.40 | read $0.120 | Serverless |
| Fireworks AI | $0.600 | $2.40 | read $0.120 | Serverless |
| Venice AI | $0.625 | $3.13 | read $0.1875 | Serverless |
| OpenRouter | - | - | - | ServerlessPartial |
| Together AI | - | - | - | ServerlessPartial |
Available via routers & gateways(1)
Capabilities
ReasoningJSON / Tool use
Benchmark peer barsfor Coding
SWE-bench VerifiedRank 72 of 104
LiveCodeBenchRank 11 of 79
Benchmark scores(10)
Scores are benchmark-specific and are direction-aware: the same numeric gap can mean very different outcomes across suites. Use the leaderboard context and this model's provider route to decide whether the winning margin is meaningful for your workload.
| Benchmark | Score | Version | Evaluation | Source |
|---|---|---|---|---|
| 56.4 | Terminal-Bench 2.1Observed 2026-10-07 | — | ||
| 70.7 | From official NVIDIA HuggingFace model cardObserved 2026-10-07 | — | ||
| 67.7 | SWE-bench Multilingual (official NVIDIA HF model card)Observed 2026-10-07 | — | ||
| 44.4 | From official NVIDIA HuggingFace model cardObserved 2026-10-07 | — | ||
| 89.0 | v6Observed 2026-10-07 | — | ||
| 87.0 | no tools; GPQA subset not stated (official NVIDIA HF model card)Observed 2026-10-07 | — | ||
| 26.7 | HLE no tools (official NVIDIA HF model card)Observed 2026-10-07 | — | ||
| 37.4 | HLE with tools (official NVIDIA HF model card)Observed 2026-10-07 | — | ||
| 86.8 | From official NVIDIA HuggingFace model cardObserved 2026-10-07 | — | ||
| 63.8 | From official NVIDIA HuggingFace model cardObserved 2026-10-07 | — |
Migration checks
No linked migration route is available for this model yet.