Weighted by its models' usage on Jul 9, 2026, the last day one was in OpenRouter's top 50 · NVIDIA sells no API of its own: its models at the providers' median price
About NVIDIA
The NVIDIA token price index is $0.200 per 1M tokens today, unchanged since yesterday, unchanged over the past week and unchanged over the past month. Since Feb 18, 2026 it has ranged from $0.0200 to $0.200. It is the usage-weighted average of the posted prices of NVIDIA's model at the mix of tokens agents send (87% cache read, 12% input, 1% output), each at the median price of the providers that serve it (NVIDIA sells no API of its own). It is calculated once a day, for each UTC day, shortly after the day ends.
NVIDIA, the chip company headquartered in Santa Clara, California, founded in 1993 and led by co-founder Jensen Huang, makes the Nemotron models and publishes their weights, training techniques and much of their training data. Nemotron 3 Ultra, released on June 4, 2026, has 550 billion parameters, 55 billion of them active; its model card puts it under the OpenMDW 1.1 license, which permits dealing in the model "without restriction", subject to its conditions. NVIDIA does not price its models per token: its API endpoints are free to members of its developer program "for prototyping", and production use of its NIM software needs an NVIDIA AI Enterprise license, priced by the number of GPUs. On October 28, 2026, NVIDIA itself served the free versions of Ultra, 3.5 Super VL and 3.5 Lightning on OpenRouter (Ultra's free endpoint), and other providers sold paid ones.
Models and prices
| Model | Input | Cache read | Output | At the mix |
|---|
| Nemotron 3 Ultra | $0.60 | $0.12 | $2.40 | $0.200 |
| Nemotron 3 Nano 30B A3B | $0.06 | $0.012 | $0.24 | $0.0200 |
USD per 1M tokens, the providers' median on OpenRouter, read on October 28, 2026, that the index uses for Ultra on October 27, 2026 and used for Nano from February 18 to June 28, 2026; "At the mix" prices 87% of tokens as cache reads, 12% as input and 1% as output, as the index does. On October 28, 2026, of three standard prices for Ultra, DeepInfra's $0.50/$2.20, Baseten's $0.60/$2.40 and Venice's $0.625/$3.125, Baseten's was the middle one at the mix. Of two for Nano, the lower at the mix was Nebius's; Novita listed $0.05/$0.20 but no cache price, so its cache reads count at its input price.
What this index measures
Each day's value averages the prices of NVIDIA's paid models, weighted by each model's tokens in OpenRouter's daily rankings, each priced at the mix of tokens agents send: 87% cache reads, 12% input and 1% output, as OpenRouter measured in agentic use. With no NVIDIA per-token price, the index takes each provider's standard price for the model on OpenRouter, without flex, fast, priority or regional tiers, ranks them at the mix and uses the middle one, or the lower of the middle two. Free versions are not counted.
On a day when no paid NVIDIA text model is among OpenRouter's 50 most-used, as on 247 of the 252 days from NVIDIA's first value on February 18, 2026 to October 27, 2026, the index keeps the model shares of the last day one was and prices them at each day's own provider prices. That day was July 9, 2026, when Nemotron 3 Ultra was the only one, so every value since is Ultra's price at the mix, $0.200 on October 27. Days before the live index began on October 28, 2026 are priced at the provider prices read that day, so the history shows which model was used, not price changes.
What moves NVIDIA's index
- Which paid model is in the top 50. On June 29, 2026, the index rose tenfold, to $0.200 from $0.0200, when the paid Ultra entered the top 50 and replaced the Nano shares carried since February 20; Baseten's Ultra prices are exactly ten times Nebius's for Nano.
- Free versions. From February 18 to October 27, 2026 they carried 99.9% of NVIDIA's tokens in OpenRouter's top 50: on October 27, the free Ultra, 3.5 Super VL and 3.5 Lightning were 14th, 21st and 31st among the models in OpenRouter's daily rankings, with 907 billion tokens between them. On July 9, the paid Ultra had 23.4 billion, 7.8% of NVIDIA's ranked tokens.
- Provider prices. A provider joining, leaving or repricing on OpenRouter can move the median with no change by NVIDIA.
- A newer paid model. The carried shares end when a paid NVIDIA text model is back in the top 50; the providers' median for Nemotron 3.5 Super VL was $0.0404 at the mix on October 28, 2026, about a fifth of Ultra's.
Price history
- January 1 to February 18, 2025: Llama 3.1 Nemotron 70B Instruct was in OpenRouter's top 50 on 16 days, with no price on record, as OpenRouter no longer lists it.
- December 15, 2025: NVIDIA announced the Nemotron 3 family and released Nano the same day.
- February 18, 2026: the paid Nano, 44th among the models in OpenRouter's rankings, gave the index its first value, $0.0200.
- March 11, 2026: Nemotron 3 Super released; to October 27, 2026, only its free version made the top 50.
- June 4, 2026: Nemotron 3 Ultra released.
- June 29, 2026: the paid Ultra entered the top 50 and the index rose to $0.200; July 9 was the last of its three days there.
- August 11 and October 15, 2026: Nemotron 3.5 Lightning and Nemotron 3.5 Super VL released; to October 27, 2026, only their free versions made the top 50.
FAQ
How much do Nemotron models cost per million tokens? On October 28, 2026, the providers' median on OpenRouter was $0.60 input and $2.40 output for Nemotron 3 Ultra; for Nano, the index uses Nebius's $0.06 and $0.24, the lower of its two providers at the index's mix (Novita listed $0.05 and $0.20 but no cache price).
Why has NVIDIA's index stayed at $0.200 since June 29, 2026? Every day since is priced as Nemotron 3 Ultra alone, NVIDIA's only paid model in the top 50 on June 29, June 30 and July 9, at the provider prices read on October 28, 2026.
How often is it updated? Once a day: each UTC day's value is calculated in the first update after that day ends, from the day's OpenRouter volumes and the providers' prices posted then. Updates are scheduled every 2 hours but can start hours late.
Sources
- NVIDIA: Nemotron
- NVIDIA Research: Nemotron 3 family, December 15, 2025
- NVIDIA Research: Nemotron 3 Ultra, June 4, 2026
- NVIDIA: NIM general FAQ
- Hugging Face: NVIDIA models
- OpenRouter: Nemotron 3 Ultra endpoints
- OpenRouter: Inferred 03, Cache is Crucial, October 8, 2026, the token mix
- OpenRouter: rankings, usage weights, CC BY 4.0
Text reviewed Oct 28, 2026; the figures in it are as of that date. The figures at the top of this page update once a day.
Cite as: Compute Screener, “NVIDIA (Nemotron) token price index”, computescreener.com/tokens/nvidia. The figures are free to reuse under CC BY 4.0 with credit to Compute Screener.
