Nemotron 3 Ultra NVFP4
NVIDIAnemotron-3-ultra-nvfp4
Released Jun 4, 2026
Providers
3
Context
1.0M
Input
$0.50/M
Output
$2.20/M
BAA-eligible on 1 of 3 providersPHI requires a signed BAA from Concentrate and HIPAA-enabled organization.
Aliases:nemotron-3-ultranvidia-nemotron-3-ultra-nvfp4
NVIDIA's frontier-scale hybrid Latent-MoE model (550B total, 55B active) for the most demanding agentic, reasoning, and long-context workloads across code, math, and science. Trained with an NVFP4 recipe.
deepinfra/nemotron-3-ultra-nvfp4Quantfp4
- Input
- $0.50/M
- Output
- $2.20/M
- Context
- 262K
- Max out
- 16K
- Cache read
- $0.10/M
- Cache write
- $0.50/M
- Thinking out
- $2.20/M
Fireworks AIDeprecating on Fireworks AI
fireworks/nemotron-3-ultra-nvfp4QuantUnspecified
- Input
- $0.60/M
- Output
- $2.40/M
- Context
- 262K
- Max out
- 66K
- Cache read
- $0.12/M
- Cache write
- $0.60/M
- Thinking out
- $2.40/M
Flex
nebius/nemotron-3-ultra-nvfp4Quantfp4
- Input
- $1.00/M
- Output
- $3.00/M
- Context
- 1.0M
- Max out
- 66K
- Caching
- Not offered
- Thinking out
- $3.00/M
FlexPriority / Fast