// pricing · September 2026
NVIDIA API Pricing
Every NVIDIA model's list price per 1M tokens — input, output and cached input — with context windows, LiveBench scores and what a real workload costs per month.
6 NVIDIA models are listed, with a median list price of $0.080 per 1M input tokens and $0.200 per 1M output tokens. Paid prices run from $0.087 (Nemotron 3 Nano 30B A3B) to $1.05 (Nemotron 3 Ultra) per 1M tokens blended. 5 have a free listing.
- Models
- 6
- 5 with a free listing
- Median input / 1M
- $0.080
- paid models
- Median output / 1M
- $0.200
- paid models
- Cheapest paid
- $0.087
- Nemotron 3 Nano 30B A3B
- Highest LiveBench
- 67.4
- Nemotron 3 Ultra
NVIDIA model prices
List prices in USD per 1M tokens. “Blended” is a 3:1 input:output mix; the per-month column prices the selected workload using each model's cached-input rate where it publishes one.
Showing 6 of 6 models · per-month column: 8K in / 600 out × 100K requests, 50% of input cached
| nvidia/nemotron-3-nano-omni-30b-a3b-reasoning:free | Free | Free | — | Free | 256K | — | $0 |
|---|---|---|---|---|---|---|---|
| nvidia/nemotron-3-nano-30b-a3b | $0.050 | $0.200 | $0.030 | $0.087 | 262K | — | $44.00 |
| nvidia/nemotron-3.5-lightning | $0.070 | $0.200 | $0.040 | $0.103 | 262K | — | $56.00 |
| nvidia/nemotron-3-super-120b-a12b | $0.080 | $0.450 | — | $0.172 | 262K | — | $91.00 |
| nvidia/nemotron-3.5-content-safety | $0.200 | $0.200 | — | $0.200 | 131K | — | $172.00 |
| nvidia/nemotron-3-ultra-550b-a55b | $0.600 | $2.40 | $0.120 | $1.05 | 262K | 67.4 | $432.00 |
NVIDIA pricing FAQ
What is the cheapest NVIDIA model?
Nemotron 3 Nano 30B A3B is the cheapest paid NVIDIA model, at $0.050 per 1M input tokens and $0.200 per 1M output tokens. 5 NVIDIA models also have a free listing, usually with tighter rate limits.
How much does the most capable NVIDIA model cost?
The highest LiveBench overall score among NVIDIA models belongs to Nemotron 3 Ultra (67.4), priced at $0.600 input and $2.40 output per 1M tokens. A RAG assistant at 100K requests a month would list at about $432.00.
Why do output tokens cost more than input tokens?
Generating a token takes a full forward pass per token, while input is processed in parallel. Across the paid NVIDIA models listed here the median output price is $0.200 per 1M against $0.080 for input — reasoning tokens bill as output too.
Which NVIDIA models support prompt caching?
3 of 6 publish a separate cached-input price, which bills repeated prompt prefixes at a fraction of the normal input rate. Filter the table by "Caching" to see them.
How current are these prices?
List prices are pulled from OpenRouter’s public models API and refreshed every 15 minutes. Batch discounts and enterprise pricing are not included.
Other providers
What these prices include
- List prices from OpenRouter's public models API, which mirrors provider pricing. Batch discounts and negotiated enterprise rates are not included.
- Variants —
:freeand:batchlistings are folded into their base model; its model page lists them. - Per-month cost is a floor: tokens you describe times list price. Cache writes and reasoning-token volume depend on your workload.