Provider pricing
Nvidia AI API pricing
Nvidia has 38 priced models in the public snapshot. For 1,000 input and 500 output tokens, Nemotron Nano 9b v2 (NVIDIA) has the lowest recorded request cost at $0.00012, with $0.04 input and $0.16 output per million tokens.
Best for
Teams running on NVIDIA GPUs, agentic workloads needing long context, or self-hosting via NIM containers.
Strengths
- Nemotron 3 Super: hybrid Mamba-Transformer MoE — 0.2× throughput overhead, 262K context
- TensorRT-LLM optimized — top-tier latency on H100/H200 hardware
- Strong agentic + tool-use performance, NeMo Guardrails ready
- Nemotron Nano 9B v2 at $0.10/M input — competitive with the cheapest tier
Tracked Nvidia models
| Model | Input $/M | Output $/M | Context | Tier | Best for |
|---|---|---|---|---|---|
| Nemotron 3.5 Super VL | $0.150 | $0.500 | 1,048,576 | Budget | Open vision-language model |
| Nemotron 3 Super 120B | $0.300 | $0.800 | 262,000 | Reasoning | Hybrid Mamba-Transformer MoE — agentic reasoning, long context |
| Llama 3.3 Nemotron Super 49B | $0.400 | $0.400 | 128,000 | Flagship | NVIDIA-tuned Llama for tool use and instruction following |
| Nemotron Nano 9B v2 | $0.100 | $0.300 | 128,000 | Budget | Cheapest fast Nemotron — high-throughput agents and RAG |
| Llama 3.1 Nemotron 70B Instruct | $0.400 | $0.400 | 128,000 | Flagship | NVIDIA-tuned Llama 70B with strong arena scores |
| Nemotron Mini 4B Instruct | $0.080 | $0.080 | 4,096 | Budget | Tiny on-device Nemotron for edge inference |
| Nemotron 4 Ultra | $0.800 | $2.40 | 256,000 | Flagship | Frontier Nemotron MoE for agentic reasoning and tool use |
Last updated: · Source: No official pricing URL recorded · Methodology
Frequently asked questions
How much does the Nvidia API cost?
Nvidia has 38 priced models in the public snapshot. For 1,000 input and 500 output tokens, Nemotron Nano 9b v2 (NVIDIA) has the lowest recorded request cost at $0.00012, with $0.04 input and $0.16 output per million tokens.
Which Nvidia model is cheapest?
Nemotron Nano 9b v2 (NVIDIA) is the lowest-cost recorded model for the 1,000-input/500-output example. Prices are $0.04 input and $0.16 output per million tokens; different workloads may change the ranking.
Does Nvidia have cached or batch prices?
Cached input: not separately recorded. Batch: not recorded in this dataset. Eligibility and cache-write charges depend on the provider; these discounts are not automatically stacked.
How often are provider prices verified?
The live pricing table synchronizes stored model records daily; public file snapshots update when the site is rebuilt. Synchronization is not independent provider verification. Snapshot updated 2026-10-31; only explicit per-model dates are verification records.