Measured: Nemotron 3.5 Super VL free vs paid row on OpenRouter

On 2026-10-21 the paid Super VL row is $0.15 / $0.50 per million tokens and the free row is $0. Bolt’s GGUF repo has more downloads than the two Super VL weight repos combined.

2026-10-21

On 2026-10-21 the public lists show two rows for Nemotron 3.5 Super VL on OpenRouter, one priced and one at zero. The same pass records Hugging Face downloads for NVFP4, BF16, and the Bolt GGUF. These are prices and counters, not a speed test.

The check is a measured Nemotron 3.5 Super VL free row versus paid row. OpenRouter prices are USD per token; the table multiplies by 1,000,000. Download figures are the Hugging Face downloads field.

Two model rowsPer million tokensThree repo counts

The two OpenRouter rows

Row	Model ID	context_length	Input / million	Output / million
Paid	`nvidia/nemotron-3.5-super-vl`	1048576	$0.15	$0.50
Free	`nvidia/nemotron-3.5-super-vl:free`	1000000	$0	$0

Raw per-token strings on the paid row: prompt 0.00000015 and completion 0.0000005. Times 1,000,000 that is $0.15 and $0.50. The same object has input_cache_read 0.00000002, or $0.02 per million tokens. The free row sets both prompt and completion to 0 and omits a cache-read price.

Three Hugging Face repos

Repo	downloads	likes	createdAt
`nvidia/NVIDIA-Nemotron-3.5-Super-VL-120B-A12B-NVFP4`	11459	188	2026-10-12
`nvidia/NVIDIA-Nemotron-3.5-Super-VL-120B-A12B-BF16`	18759	82	2026-10-12
`nvidia/NVIDIA-Nemotron-3.5-Bolt-74B-A9B-24GB-GGUF`	87789	105	2026-10-07

11459Super VL NVFP4 downloads

18759Super VL BF16 downloads

87789Bolt GGUF downloads

11459 plus 18759 is 30218. The Bolt GGUF repo shows 87789 downloads, which is 57571 more than those two Super VL repos combined. The GGUF repo was created on 2026-10-07; the two Super VL repos were created on 2026-10-12. The formats are NVFP4, BF16, and GGUF. The comparison is the downloads field.

Likes are a separate field: 188 on NVFP4, 82 on BF16, and 105 on the Bolt GGUF. The search used was https://huggingface.co/api/models?search=Nemotron-3.5.

The figures in the tables come from the two public APIs on 2026-10-21. The photo is not a benchmark chart.

Pull the same numbers again

1
Open the model list
GET https://openrouter.ai/api/v1/models and find nvidia/nemotron-3.5-super-vl plus the :free id.
2
Scale to one million tokens
Multiply pricing.prompt and pricing.completion by 1,000,000. On the paid row also read input_cache_read.
3
Copy three repos
GET the Hugging Face search for Nemotron-3.5 and copy downloads, likes, and createdAt for NVFP4, BF16, and the Bolt GGUF.
4
Add, then compare
11459 + 18759 = 30218. Set that next to 87789 and keep the repo ids and the fetch date in the sentence.

https://openrouter.ai/api/v1/models
https://huggingface.co/api/models?search=Nemotron-3.5

Fits: Writing down the published price and download counters before choosing the free row, the paid row, or a weight file.
Outside this note: Using the page as a latency or throughput test, or treating downloads as a quality ranking.

Is the free row a different weight?

Both rows share a hugging_face_id and the Nemotron 3.5 Super VL description. They differ in context_length, price, and the top_provider context and max completion fields.

What does 87789 versus 30218 mean?

87789 is the Bolt GGUF repo’s downloads. 30218 is NVFP4 plus BF16. The gap of 57571 is that counter. The three repos use different formats and creation dates.

Will $0.15 stay the price?

This page stores the 2026-10-21 read of the two URLs. Later prices are whatever pricing those endpoints return then.

Is context 1048576 or 262144?

On the paid row the model field context_length is 1048576, and top_provider.context_length is 262144. On the free row both fields are 1000000. Name the field when you cite it.

Put the two rows on one page

To point at the prices and repo ids with someone else, open one shared board. Steps: create and join a space.