Nemotron 3 Ultra API pricing
Nvidia · Mid-range price tier · OpenMDW 1.1 · released 2026-06. Prices in USD per 1 million tokens. Data snapshot: 2026-11-12.
A support ticket workload of 100K requests a month costs $210.00 on Nemotron 3 Ultra; the next cheaper model in the Mid-range price tier for this workload, Qwen3.6 27B, does it for $208.00. The price we quote for Nemotron 3 Ultra has changed 11 times since 2026-08-15, most recently on 2026-10-11 when the quoted host changed (+10.1%); it is now the 21st cheapest of 113 models in the Mid-range price tier. Buying from DeepInfra rather than at the price we quote lowers the token price by about 9%, for a quantized fp4 build. It holds the FinOps Friendly badge: top-40% Arena quality, top-30% Efficiency, not among the dearest 30%, and no stability flag.
- Input: $0.60 per 1M tokens
- Output: $2.40 per 1M tokens
- Batch API (in/out): Not published
- Prompt-cache read: $0.12 per 1M tokens
- Context window: 262K tokens
- Arena ELO: 1445 as of 2026-11-08
Market position
On blended price ($1.86 per 1M tokens at 30% input, 70% output) it is cheaper than 81% of models in the Mid-range price tier: 21st cheapest of 113, tied with one other. Across every price tier it is cheaper than 40% of priced text models in the catalogue: 181st cheapest of 303, tied with one other. Its Arena ELO of 1445 ranks 44th of the 160 models that carry one, tied with 2 others, above 72% of them. With an Arena score of 1445, Nemotron 3 Ultra clears OptimNow's Standard working quality threshold: Standard (Arena 1420+), used for meeting summary, invoice processing and call summary. Each threshold is a distance below the best ranked model, which scores 1530 on Arena scores as of 8 Nov 2026.
Licence, size and capabilities
Nemotron 3 Ultra is open weights, which is not the same as open source: the weights are published, but under OpenMDW 1.1, a licence that restricts some uses. Read it before self-hosting commercially. Its published size is 550B parameters. Listed capabilities: Text, Reasoning, Agents. Filling its 262K context window once costs $0.1572 in input tokens at list price.
Pricing by host
A host is a company that runs the model and sells it by the token: the model maker, a cloud platform like Amazon Bedrock, Azure or Vertex, or an independent host. One model, several hosts, and often several prices. This model is sold by 4 different hosts. The price we quote elsewhere on this page is marked. We quote one host's rate so that models stay comparable; it is not a recommendation to buy there.
Cheapest host: DeepInfra at $0.50 input / $2.20 output per 1M tokens, 9% below the price we quote on the blended price (30% input, 70% output). It serves a quantized fp4 build, which is not the same product as the full-precision model. A support ticket workload at 100K requests a month costs $185.00 there against $210.00 at the price we quote, a difference of $25.00 a month. The dearest, Venice, charges 1.4x the cheapest.
| Host | Input $/1M | Output $/1M | Precision |
|---|---|---|---|
| DeepInfra | $0.50 | $2.20 | fp4 |
| KT | $0.50 | $2.20 | not declared |
| BaseTen (price we quote) | $0.60 | $2.40 | fp4 |
| Venice | $0.625 | $3.125 | fp8 |
List prices as published on 2026-11-12, via OpenRouter. Negotiated rates, committed-use discounts and provisioned capacity are not shown. These hosts do not all serve the model at the same numeric precision, so the cheapest is not a like-for-like substitute for the dearest.
Monthly cost at 100K requests, by business use case
From the after-discounts figure (prompt caching on the cached input share, batch API for async workloads, only where Nvidia publishes those rates), which assumes you implement both, to the list price. One figure means no discount is published.
| Use case | Tokens (in/out) | Monthly (after discounts – list) |
|---|---|---|
| Support Ticket | 1,500 / 500 | $167 – $210 |
| Knowledge Q&A | 2,000 / 800 | $245 – $312 |
| Meeting Summary | 10,000 / 1,200 | $840 – $888 |
| Marketing Content | 2,500 / 1,800 | $558 – $582 |
| Coding Task | 3,000 / 2,000 | $588 – $660 |
| Invoice Processing | 1,500 / 600 | $212 – $234 |
| Call Summary | 2,000 / 700 | $278 – $288 |
| Agent Workflow | 6,000 / 3,000 | $878 – $1.1K |
Price history
Nemotron 3 Ultra has been in our daily price archive since tracking began on 2026-08-15, when it listed at $0.60 input / $3.60 output per 1M tokens. The price we quote has changed 11 times since then, 10 of them a change of quoted host rather than a repricing. The latest change, on 2026-10-11, was +10.1% on the blended price (30% input, 70% output), to $0.60 input / $2.40 output (quoted host changed, not a repricing). Net of all changes, the blended price is -31.1% against 2026-08-15.
| Date | Old price (in / out) | New price (in / out) | Blended change |
|---|---|---|---|
| 2026-10-11 | $0.50 / $2.20 | $0.60 / $2.40 | +10.1% (quoted host changed, not a repricing) |
| 2026-10-04 | $0.60 / $2.40 | $0.50 / $2.20 | -9.1% (quoted host changed, not a repricing) |
| 2026-10-03 | $0.50 / $2.20 | $0.60 / $2.40 | +10.1% (quoted host changed, not a repricing) |
| 2026-10-02 | $0.60 / $2.40 | $0.50 / $2.20 | -9.1% (quoted host changed, not a repricing) |
| 2026-09-19 | $0.625 / $3.125 | $0.60 / $2.40 | -21.7% (quoted host changed, not a repricing) |
6 earlier changes are not listed.
Alternatives to Nemotron 3 Ultra
None of the 4 comparable models below costs less than Nemotron 3 Ultra on the blended price.
- Next cheaper model in the Mid-range price tier: Nova 2 Lite (Amazon) at $1.84 blended, 1% less.
- Next more expensive: Qwen3.5-27B (Alibaba) at $1.898 blended, 2% more.
The nearest models in the Mid-range price tier on Arena ELO, each with its blended price and score difference against Nemotron 3 Ultra.
- DeepSeek V4 Pro 0813 (DeepSeek): $0.111 in / $4.44 out per 1M, Arena ELO 1445 (69% more expensive, same Arena ELO)
- Kimi K2.5 (Moonshot): $0.532 in / $3.325 out per 1M, Arena ELO 1445 (34% more expensive, same Arena ELO)
- GPT-6.1 Sol (OpenAI): $2.00 in / $10.00 out per 1M, Arena ELO 1446 (309% more expensive, 1 ELO point higher)
- GLM 5 (Zhipu): $1.00 in / $3.20 out per 1M, Arena ELO 1446 (37% more expensive, 1 ELO point higher)
- Cheapest within 15 Arena points: Gemma 4 26B A4B (Google), 95% cheaper, 11 ELO points lower, out of 39 models in that band.
- Highest Arena ELO within 20% of this price: DeepSeek V4 Pro 0423 (DeepSeek), 13% cheaper, 6 ELO points higher, out of 21 models in that price band.
Other models from Nvidia
Nvidia has 5 other models in the catalogue. Prices per 1M tokens, input then output.
- Nemotron 3.5 Lightning: $0.07 / $0.20
- Nemotron 3 Nano 30B A3B: $0.06 / $0.24
- Nemotron 3.5 Content Safety: $0.20 / $0.20
- Nemotron 3 Super: $0.075 / $0.40
- Nemotron 3.5 Super VL: $0.15 / $0.47
All 6 Nvidia models and prices
Common questions about Nemotron 3 Ultra pricing
- How much does Nemotron 3 Ultra cost?
- Nemotron 3 Ultra costs $0.60 per 1M input tokens and $2.40 per 1M output tokens at BaseTen, the host whose rate we quote. BaseTen serves a quantized fp4 build, which is not the same product as the full-precision model. Prices as listed on 2026-11-12.
- Which host is cheapest for Nemotron 3 Ultra?
- Cheapest host: DeepInfra at $0.50 input / $2.20 output per 1M tokens, 9% below the price we quote on the blended price (30% input, 70% output). It serves a quantized fp4 build, which is not the same product as the full-precision model. A support ticket workload at 100K requests a month costs $185.00 there against $210.00 at the price we quote, a difference of $25.00 a month. The dearest, Venice, charges 1.4x the cheapest. Host rates as listed on 2026-11-12.
- Does Nemotron 3 Ultra offer batch or cached pricing?
- Yes, for prompt caching only. Cache reads cost $0.12 per 1M tokens, as listed on 2026-11-12. No batch pricing is published.
- What would Nemotron 3 Ultra cost per month?
- With the Support Ticket profile (1,500 input and 500 output tokens per request), 100,000 requests a month cost about $210 on Nemotron 3 Ultra at the list price quoted on 2026-11-12.
- Can Nemotron 3 Ultra be self-hosted?
- Yes, within the terms of its licence. Nemotron 3 Ultra is open weights, which is not the same as open source: the weights are published, but under OpenMDW 1.1, a licence that restricts some uses. Read it before self-hosting commercially.
- Has the price of Nemotron 3 Ultra changed?
- Nemotron 3 Ultra has been in our daily price archive since tracking began on 2026-08-15, when it listed at $0.60 input / $3.60 output per 1M tokens. The price we quote has changed 11 times since then, 10 of them a change of quoted host rather than a repricing. The latest change, on 2026-10-11, was +10.1% on the blended price (30% input, 70% output), to $0.60 input / $2.40 output (quoted host changed, not a repricing). Net of all changes, the blended price is -31.1% against 2026-08-15.
Its weights are published, which changes what you may run yourself: the cost gap between open weights and proprietary APIs.
These are list prices. Your real cost also depends on caching and batching, negotiated discounts, marketplace agreements, and the contracts you already have. OptimNow audits exactly that.
Build the business case · Get a FinOps review
LLM pricing · Models by maker · Cloud compute pricing · Compare models · Price barometer · FinOps guides · API and MCP server