Skip to content
NINJA

Nemotron 3 Ultra

Released by NVIDIA on 4 Jun 2026.

API id nvidia/nemotron-3-ultra-550b-a55b

Input price
$0.60 / 1M tokens
Output price
$2.40 / 1M tokens
Context window
262K tokens
Longest reply
183K tokens

About this model

NVIDIA Nemotron 3 Ultra is an open frontier-reasoning and orchestration model from NVIDIA, with 55B active parameters out of 550B total (MoE). Built on a hybrid Transformer-Mamba mixture-of-experts architecture, it...

Description from the model’s public catalogue listing.

Capabilities

Reads
Text
Writes
Text
Reasoning mode
Yes
Tool calling
Yes
Open weights
Yes
Cached input
$0.12 / 1M tokens

What it would cost

At list price, without caching or batch discounts.

1,000 chat replies2K tokens in, 500 out each
$2.40
Summarise 100 long reports50K tokens in, 1K out each
$3.24
A coding agent’s working day20M tokens in, 1M out
$14.40
Price your own workload on every model

Price history

Per 1M tokens, input then output. Tracked since 30 Sept 2026.

11 Oct 2026
$0.50 → $0.60+20%$2.20 → $2.40+9%
2 Oct 2026
$0.60 → $0.50-17%$2.40 → $2.20-8%
2 Oct 2026
$0.50 → $0.60+20%$2.20 → $2.40+9%
2 Oct 2026
$0.60 → $0.50-17%$2.40 → $2.20-8%

Alternatives at a similar price

From other labs, priced within about 40% of Nemotron 3 Ultra, with a comparable context window.

More from NVIDIA

See all