Nemotron 3 Ultra vs Nemotron 3.5 Super VL
Nemotron 3 Ultra scores higher on the Models at Work leaderboard (55 vs 50 out of 100). On independent tests alone, Nemotron 3 Ultra is ahead (50 vs 32). Nemotron 3.5 Super VL has no list price we could verify, so the two cannot be compared on cost. On the Artificial Analysis Intelligence Index, one of the independent tests, Nemotron 3.5 Super VL leads 14 to 12.
| Nemotron 3 Ultra · NVIDIA | Nemotron 3.5 Super VL · NVIDIA | |
|---|---|---|
| Leaderboard rank | #24 | #29 |
| Score (0–100) | 55 | 50 |
| Kind of work | Run it yourself | Run it yourself |
| Price per 1M tokens | $0.50 to read · $2.20 to write | Not verified |
| Reads at once | 262K tokens (about 197,000 words) | 1M tokens (about 750,000 words) |
| Where you can get it | NVIDIA build (build.nvidia.com) | NVIDIA NIM, OpenRouter |
| Can a company host it itself | Yes (open weights) | Yes (open weights) |
| Artificial Analysis Intelligence Index | 12 | 14 |
| Arena Text score (nvidia-nemotron-3-ultra-550b) | 1426 (rank 120) | Not measured |
| Scale MCP Atlas (thinking) | 63.10 | Not measured |
| Humanity's Sixth Sense (Scale) | Not measured | 24.70 |
| Best for | Companies that cannot use a Chinese model, High-volume routine tasks on your own NVIDIA servers, Answering questions from company documents | Pulling fields out of documents and images in-house, Work where data cannot leave your servers |
Cost of 1,000 jobs
| 1,000 of these | Nemotron 3 Ultra | Nemotron 3.5 Super VL |
|---|---|---|
| Support repliesthe customer message, the account notes and two help articles read; a 300-word reply written | $2.38 | No list price |
| Documents processeda two-page invoice or form read; the fields pulled out | $1.66 | No list price |
| Calls summariseda 40-minute call transcript read; a summary and next steps written | $6.32 | No list price |
| Contracts revieweda 40-page contract read against a checklist; a memo of issues written | $18 | No list price |
| Drafts writtena brief and one example read; a 600-word email or proposal section written | $2.51 | No list price |
Worked out from each vendor’s list price and a fixed size for each job. A product built on a model charges its own price on top. See what every model costs for the same jobs.
Choose Nemotron 3 Ultra for
- Companies that cannot use a Chinese model
- High-volume routine tasks on your own NVIDIA servers
- Answering questions from company documents
The open model for a company that needs a US-built one on its own NVIDIA hardware; at 12 on Artificial Analysis's index it trails the Chinese open models by a wide margin.
Caveats: Its score on Artificial Analysis's index has fallen with each rebuild of that index and now sits at 12, well behind GLM-5.3 (32). NVIDIA publishes no per-use price; the $0.50 in / $2.20 out per million tokens here is a reseller's (OpenRouter). It can read about a quarter as much text at once as most models on this board, despite some resellers advertising more. One test found replies taking up to 40 seconds, too slow for live chat.
Choose Nemotron 3.5 Super VL for
- Pulling fields out of documents and images in-house
- Work where data cannot leave your servers
An open NVIDIA model that reads documents and images on your own hardware; 14 on Artificial Analysis's index says use it to pull information out, not to reason.
Caveats: There is no per-use vendor price: it comes as downloadable weights or as NVIDIA NIM, NVIDIA's packaged version for running on your own servers. The licence (OpenMDW 1.1) is as recorded in the release tracker; it was not re-read. NVIDIA's own speed and accuracy figures are not counted.
What people running them say
Nemotron 3 Ultra: Little unprompted talk from users compared with the Chinese open models. Artificial Analysis's launch note called it 'leading US open weights intelligence' at 'over 300 tokens per second' but still behind Kimi K2.6. A test of planning, WebBrain, run through the reseller OpenRouter, scored it 81 out of 100 but found its slowest one-in-twenty requests took 40.6 seconds or more, 'hard to imagine' for live use. Most discussion is about how the model is built rather than day-to-day results at writing software.
Questions people ask
Which is better, Nemotron 3 Ultra or Nemotron 3.5 Super VL?
Nemotron 3 Ultra scores higher on the Models at Work leaderboard (55 vs 50 out of 100). On independent tests alone, Nemotron 3 Ultra is ahead (50 vs 32). Nemotron 3.5 Super VL has no list price we could verify, so the two cannot be compared on cost. On the Artificial Analysis Intelligence Index, one of the independent tests, Nemotron 3.5 Super VL leads 14 to 12.
When should I choose Nemotron 3 Ultra over Nemotron 3.5 Super VL?
Choose Nemotron 3 Ultra for companies that cannot use a chinese model, high-volume routine tasks on your own nvidia servers, answering questions from company documents. The open model for a company that needs a US-built one on its own NVIDIA hardware; at 12 on Artificial Analysis's index it trails the Chinese open models by a wide margin.
When should I choose Nemotron 3.5 Super VL over Nemotron 3 Ultra?
Choose Nemotron 3.5 Super VL for pulling fields out of documents and images in-house, work where data cannot leave your servers. An open NVIDIA model that reads documents and images on your own hardware; 14 on Artificial Analysis's index says use it to pull information out, not to reason.
Scores come from the Models at Work leaderboard: Score is 0 to 100 and computed, not typed. 40% is independent testing (Artificial Analysis, Arena, Scale, Epoch, Vals and others; each test is scaled against the best score on this board, and tests of real work, such as finished freelance projects or professional tasks, count double the exams). 30% is price: the vendor's list price for a typical mix of reading and writing, on a fixed scale from $0.05 to $60 per million tokens, so a model that costs a fifth as much gains real ground. 15% is what people report after running it, from Signals and linked threads. 15% is fit: how much text it can take in at once and where a business can get it (the products and clouds that carry it), with a small credit for open weights. A missing component drops out and the rest are reweighted. A model with no independent test yet is provisional and ranks below every measured one. Within three points is a tie. Vendor-published figures are labelled and never counted. Prices are vendor list prices where published; a typical mix assumes three parts reading to one part writing.