Nvidia
Nemotron 3 Ultra
nvidia/nemotron-3-ultra-550b-a55bHugging Face ↗
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...
Only tracked on Vercel AI Gateway.
Nemotron 3 Ultra usage over time
Vercel AI Gateway share of tokens, by day.
Nemotron 3 Ultra holds 2.15% today, up from 0.44% at the start of the visible window.
Rank history
Adoption curve vs the pantheon
Nemotron 3 Ultra's share plotted against days since first tracked, overlaid on other current leaders on the same axis. The question every new-model page asks: is this one going to make it?
Citable stat
As of 2026-09-02, Nemotron 3 Ultra holds 0.0% of Vercel AI Gateway token share, down 0.01pp over 30 days. Source: Tidelines (Vercel AI Gateway).
Cost per session
Median dollars spent on one Nemotron 3 Ultra session, by coding agent and session length.
Token rate sets the price; session length and reasoning set the bill.
| Agent | 1 turn | 2–9 turns | 10–49 turns |
|---|---|---|---|
| Hermes Agent | $0.012 | $0.035 | $0.307 |
Where to run it
Nemotron 3 Ultra pricing by provider
Per-provider pricing, context, and uptime via OpenRouter. Sorted by output price.
Nemotron 3 Ultra is served by 3 tracked endpoints on OpenRouter. The cheapest output price is $2.20 per million tokens from DeepInfra, and the most expensive is $3.13 from Venice. The largest context window offered is 262k tokens. Prices are list rates and refresh daily.
| Provider | Input $/M | Output $/M | Context | Uptime |
|---|---|---|---|---|
| DeepInfra | $0.50 | $2.20 | 262k | — |
| BaseTen | $0.60 | $2.40 | 203k | 100.00% |
| Venice | $0.63 | $3.13 | 256k | 100.00% |