Analysis: Nemotron 3.5 Super VL OpenRouter price vs HF downloads
Paid and free context lengths are 1048576 and 1000000. On Hugging Face, BF16 has 18759 downloads and NVFP4 has 188 likes. Neither count is a quality score.
To choose OpenRouter token billing or a Hugging Face NVFP4 or BF16 download, line up the public fields from 2026-10-21 first. This note cites no eval scores.
The listing is NVIDIA Nemotron 3.5 Super VL. The paid id is nvidia/nemotron-3.5-super-vl. The free id is nvidia/nemotron-3.5-super-vl:free. OpenRouter stores US dollars per token. The table multiplies by 1,000,000.
Paid and free rows on OpenRouter
Both rows share created Unix 1792076662 (2026-10-15T15:04:22Z) and canonical_slug nvidia/nemotron-3.5-super-vl-20261015. Paid prompt is 0.00000015, completion is 0.0000005, and input_cache_read is 0.00000002. Per 1,000,000 tokens those are $0.15, $0.50, and $0.02. The free pricing object has prompt and completion at 0 and no cache field.
| OpenRouter id | context_length | top_provider context | Input / 1,000,000 | Output / 1,000,000 | Cache read / 1,000,000 |
|---|---|---|---|---|---|
nvidia/nemotron-3.5-super-vl |
1048576 | 262144 | $0.15 | $0.50 | $0.02 |
nvidia/nemotron-3.5-super-vl:free |
1000000 | 1000000 | $0 | $0 | field absent |
The listing text calls it an open multimodal mixture-of-experts model from NVIDIA, with 12B active parameters out of 120B total, taking text, images, documents, GUI screenshots, and video. The API description ends in an ellipsis, so this page does not invent the rest. Architecture input is text, image, and video; output is text. The 120B-A12B names on Hugging Face match that listing text. This page did not remeasure parameter counts.
NVFP4 and BF16 on Hugging Face
The search string is NVIDIA-Nemotron-3.5-Super. The table is only the two official NVIDIA weight repos. downloads is a fetch count and likes is a different counter. Neither is a quality score. BF16 has more downloads. NVFP4 has more likes. The two counters point different ways.
| Hugging Face repo | downloads | likes | createdAt |
|---|---|---|---|
nvidia/NVIDIA-Nemotron-3.5-Super-VL-120B-A12B-NVFP4 |
11459 | 188 | 2026-10-12T15:32:30.000Z |
nvidia/NVIDIA-Nemotron-3.5-Super-VL-120B-A12B-BF16 |
18759 | 82 | 2026-10-12T15:32:06.000Z |
The same search includes official base weights at nvidia/NVIDIA-Nemotron-3.5-Super-VL-120B-A12B-Base-BF16: 1030 downloads, 52 likes, createdAt 2026-10-12T15:31:49.000Z. Do not add third-party conversions into the table. ggml-org/NVIDIA-Nemotron-3.5-Super-VL-120B-A12B-GGUF has 31763 downloads and 9 likes, createdAt 2026-10-15T16:01:56.000Z. lmstudio-community/NVIDIA-Nemotron-3.5-Super-VL-120B-A12B-GGUF has 26971 downloads and 2 likes, createdAt 2026-10-14T15:21:07.000Z. Both official weight repos use pipeline image-text-to-text.
OpenRouter sets hugging_face_id to nvidia/NVIDIA-Nemotron-3.5-Super-VL. This public search did not return a download count for that exact id. A direct request returned HTTP 401, so this page quotes no downloads for it.
What each number can decide
Pay per token
Use the paid row: $0.15 in and $0.50 out. Keep both 1048576 on the model and 262144 on top_provider. They arrived in the same response.
Free row
The price is $0, but context_length is 1000000, not the paid 1048576. The payload has no cache price. Do not fill one in as zero.
Download weights
NVFP4 and BF16 are two files. 18759 versus 11459 is a difference in fetches, not a quality rank.
Check the fields again later
-
1
Read both OpenRouter prices
Open
https://openrouter.ai/api/v1/models, find the two ids, and multiply USD per token by 1,000,000. -
2
Record each context_length
If paid is still 1048576 and free is still 1000000, do not write them as one window. Then read top_provider context.
-
3
Read only official NVFP4 and BF16
Search Hugging Face for
NVIDIA-Nemotron-3.5-Superand copy downloads, likes, and createdAt for those two nvidia repos. Count community GGUF repos separately.
To spread this table with someone else, open a browser space on tidemeet. The canvas and voice sit on one page, and the space does not ask for an account first.
Why do the free and paid context lengths differ?
On the 2026-10-21 model objects, paid context_length is 1048576 and free context_length is 1000000. That split is in the listing. Paid top_provider also reports 262144.
Do more downloads mean a better checkpoint?
No. BF16 has 18759 downloads and 82 likes. NVFP4 has 11459 downloads and 188 likes. The counters disagree, and some community GGUF repos have still higher download counts. This page does not treat downloads as a quality score and cites no benchmark.
Is $0.15 the price of one token?
No. The prompt field is 0.00000015 USD per token. Times 1,000,000, that is $0.15 per 1,000,000 tokens. Completion 0.0000005 is $0.50 per 1,000,000 tokens.
Is the OpenRouter Hugging Face id the NVFP4 repo?
No. hugging_face_id is nvidia/NVIDIA-Nemotron-3.5-Super-VL, which matches neither the NVFP4 nor the BF16 repo id. The public search did not return downloads for that exact id, and a direct request returned HTTP 401.