Nemotron 3 Ultra API pricing

6 providers serve Nemotron 3 Ultra. Nemotron 3 Ultra is an open-weight model (OpenMDW-1.1) with 550B total parameters, 55B active per token (mixture-of-experts), up to 1049K context. Prices below are per 1M tokens, cheapest input first (self-hosting math further down).

Refreshed 2026-10-08 · list prices from provider APIs

Input floor $0.500/1M at W&B Inference, DeepInfra and OpenRouter Output floor $2.15/1M at W&B Inference Find your workload winner →

Provider	Input $/1M	Output $/1M	Cache read $/1M	Context
W&B Inference	$0.500	$2.15output floor	$0.100	262K	Use
DeepInfra	$0.500	$2.20	$0.100	262K	Use
OpenRouter	$0.500	$2.20	$0.100	262K	Use
Together AI	$0.600	$3.60	$0.200	512K	Use
Baseten	$0.600	$2.40	$0.120	203K	Use
Nebius	$1.00	$3.00	-	1049K	Use

The cheapest way to run Nemotron 3 Ultra

On an illustrative 70M-input / 30M-output monthly workload, today's lowest listed cost is roughly $99.50/month, at W&B Inference. Input-heavy, output-heavy and cache-heavy workloads can produce different winners. Use the live mix calculator below rather than combining floors from two different providers. On output-token cost alone, a busy self-hosted deployment can work out cheaper (~$1.51/1M output on 4× MI300X). See the math below. Run your own numbers in the breakeven calculator, or read the breakeven math.

Your actual monthly cost

Uses today's list prices

Workload preset Input tokens / month (millions) Output tokens / month (millions) Input served from cache %

Cache read uses the listed cache rate where available; otherwise it falls back to normal input price. Batch, write-cache, volume and negotiated discounts are excluded.

Or run it yourself: Nemotron 3 Ultra self-hosting economics

Nemotron 3 Ultra is open-weight (OpenMDW-1.1), so the API price above competes with the GPU-hour market. 550B total parameters (MoE, ~55B active per token) needs roughly 757 GB VRAM at FP8 or 379 GB at INT4, KV-cache headroom included.

No single GPU fits Nemotron 3 Ultra at FP8 (757 GB needed), so it needs a multi-GPU node. Cheapest tracked option: 4× MI300X (192 GB each) at roughly $2.00/hr total (RunPod). Rent 4× MI300X at RunPod → INT4 quantization drops the requirement to 379 GB. Tune it in the calculator.

Breakeven estimate: a well-batched vLLM deployment on 4× MI300X ($2.00/hr) at ~736 aggregate tok/s and 50% utilization works out to roughly $1.51 per 1M output tokens, versus $2.15 via the cheapest output-token API (W&B Inference). Self-hosting wins on cost if you can keep the GPU busy. Planning estimate: throughput varies with hardware, quantization, batch size and context. Tune it in the calculator.

Related models

Model	Cheapest in $/1M	Cheapest out $/1M	Providers
Nemotron 3 Nano	$0.050	$0.200	4
Nemotron 3.5 Lightning	$0.060	$0.160	3
Nemotron 3 Super	$0.080	$0.400	3
Nemotron 3.5 Content Safety	$0.200	$0.200	1

FAQ

What is the cheapest API for Nemotron 3 Ultra?

As of 2026-10-08, the lowest input price for Nemotron 3 Ultra, $0.500 per 1M tokens, is the same list price at W&B Inference, DeepInfra and OpenRouter. The lowest output price is W&B Inference at $2.15 per 1M output tokens. The cheapest provider for a real workload depends on its input/output mix. The next-lowest input price is Together AI at $0.600, a 20% difference.

How much VRAM do you need to self-host Nemotron 3 Ultra?

Nemotron 3 Ultra has 550B parameters, so plan for roughly 757 GB of VRAM at FP8 or 379 GB at INT4/AWQ, KV-cache headroom included. That exceeds a single GPU. A typical node is 4× MI300X.

← All models · GPU rental prices · Breakeven calculator · Get this data as JSON