API pricing
NVIDIA API pricing
NVIDIA API prices run from $0.05 per million input tokens (Nemotron 3 Nano 30B A3B) to $0.50 (Nemotron 3 Ultra). Its highest-ranked model, Nemotron 3 Ultra, costs $0.50 input and $2.20 output per million tokens.
Last verified
| Cached input | ||||||
|---|---|---|---|---|---|---|
| Nemotron 3 Nano 30B A3B (open weights) | $0.05 | $0.20 | $0.025 | $0.0875 | 262K | 40.6 |
| Nemotron 3.5 Lightning (open weights) | $0.05 | $0.20 | $0.01 | $0.0875 | 262K | 40.0 |
| Nemotron 3 Super (open weights) | $0.08 | $0.45 | — | $0.17 | 262K | 40.1 |
| Nemotron 3.5 Content Safety (open weights) | $0.20 | $0.20 | — | $0.20 | 131K | — |
| Nvidia Llama 3.3 Nemotron Super 49b v1.5 (open weights) | $0.40 | $0.40 | — | $0.40 | 131K | 40.3 |
| Nemotron 3 Ultra (open weights) | $0.50 | $2.20 | $0.10 | $0.93 | 262K | 42.5 |
Prices on other platforms
The same models are often sold through clouds and resellers at different rates.
| Model | Route | Input | Output | Checked |
|---|---|---|---|---|
| Nemotron 3 Ultra | fireworks | $0.60 | $2.40 | 2026-10-10 |
| openrouter | $0.50 | $2.20 | 2026-10-10 | |
| together | $0.60 | $3.60 | 2026-10-10 | |
| Nemotron 3 Nano 30B A3B | bedrock | $0.06 | $0.24 | 2026-10-10 |
| deepinfra | $0.05 | $0.20 | 2026-10-10 | |
| openrouter | $0.06 | $0.24 | 2026-10-10 | |
| Nvidia Llama 3.3 Nemotron Super 49b v1.5 | deepinfra | $0.40 | $0.40 | 2026-10-10 |
| Nemotron 3 Super | bedrock | $0.15 | $0.65 | 2026-10-10 |
| openrouter | $0.08 | $0.45 | 2026-10-10 | |
| Nemotron 3.5 Lightning | fireworks | $0.05 | $0.20 | 2026-10-10 |
| openrouter | $0.07 | $0.20 | 2026-10-10 | |
| Nemotron 3.5 Content Safety | openrouter | $0.20 | $0.20 | 2026-10-10 |
Frequently asked questions
How much does the NVIDIA API cost?
NVIDIA API prices run from $0.05 per million input tokens (Nemotron 3 Nano 30B A3B) to $0.50 (Nemotron 3 Ultra). Its highest-ranked model, Nemotron 3 Ultra, costs $0.50 input and $2.20 output per million tokens.
Does NVIDIA discount cached input?
Yes. Cached input is billed at a lower rate on 3 of the 6 priced models; the cached rate is listed next to each model.