Nemotron 3 Ultra 550B
NVIDIA · open weights · writing benchmark
The short version
Nemotron 3 Ultra 550B sits at #79 of 130 on ToneBench, toward the value end of the board, with a writing Elo of 1394 and an overall score of 76.8 out of 100. We measured it by having it write all 9 of our real YouTube scripts, five times each, then scoring every draft blind against our own finished versions. Here is how it shook out.
Inside NVIDIA, this is the strongest writer we tested. It edges out the other 3 NVIDIA configs on the board, so if you're staying in this family for voice work, this is the one to reach for.
Among open-weights models, it comes in 19th of 47. Open weights still trail the closed frontier on pure voice fidelity, and you can see that in the gap at the top.
What it does best is Visual Cues: it ranks 66th on that dimension at 81.4, well above the board average. Its softer spot is Anti-Slop (74.3, 102nd), which is the thing to watch if that metric matters most for your use.
It was uneven across the 9 scripts. Its best run was the short explainer (80.4) and its weakest was the personal technical walkthrough / agentic workflow case study (71.2), a spread of about 9 points. Worth knowing if your writing skews toward one of those formats.
On cost, it runs about $0.011/task (estimated). DeepSeek V4 Flash (native chat alias) scores higher for less money, at $0.002/task, so on pure value this config is not the frontier. Still, it beats 15 models that cost noticeably more.
Skill profile
The shape is the story: the further a corner reaches, the stronger that dimension. Biggest edge over the board: Visual Cues (+6.8); furthest behind: Anti-Slop (-8.5).
Per-metric scores
Nine writing dimensions, each scored 0–100 and blended by the editorial weight shown. Rank is against all 130 current-ranked models. The thick bar is this model; the thin lines above and below are the current field's best model (darker) and average (lighter) on the same scale. The small ± number is its run-to-run variation.
Per-article scores
The same 9 real scripts every current-ranked model writes, scored individually. Different formats stress different skills.
Consistency: run-to-run variability
Every cell on this page is the mean of 5 independent runs per script; the ± numbers are the run-to-run standard deviation. Overall, Nemotron 3 Ultra 550B varies by ± 4.7 points between runs versus a board median of ± 3.8, so it is about as repeatable as the typical model on the board.
Its most volatile dimension is Writing Craft (± 5.5 vs a board median of ± 2.0): two runs of the same brief can land visibly different writing craft scores.
| Metric | This model | Board median | Verdict |
|---|---|---|---|
| Tone & Voice ? | 79.0 ± 3.1 | ± 2.2 | typical |
| Writing Craft ? | 76.3 ± 5.5 | ± 2.0 | swingier than most |
| Substance & Value ? | 79.0 ± 3.7 | ± 2.6 | typical |
| Flow & Emotion ? | 74.5 ± 4.3 | ± 2.6 | swingier than most |
| YouTube Structure ? | 73.9 ± 4.5 | ± 3.6 | typical |
| Hook ? | 83.4 ± 6.2 | ± 3.0 | swingier than most |
| Length Discipline ? | 67.5 ± 10.5 | ± 8.6 | typical |
| Anti-Slop ? | 74.3 ± 4.4 | ± 2.6 | swingier than most |
| Visual Cues ? | 81.4 ± 6.0 | ± 4.7 | typical |
Measured, not modeled
Numbers we log on every run and rarely talk about. None of these affect the writing scores; cost and latency are informational.
What each judge scored it
The published overall of 76.8 is the consensus of three family-disjoint judges scoring the same 45 stored drafts blind with the identical rubric. The highest and lowest judge differ by 13.9 points on its overall. That gap is wider than typical (board median 9.0): DeepSeek V4 Flash rates it noticeably higher than Claude Opus 5. Read the per-metric numbers knowing the consensus sits between two real opinions. How the panel works: methodology.
How we ran Nemotron 3 Ultra 550B
- Accessed viaOpenRouter · model page ↗
- Exact model / route id
nvidia/nemotron-3-ultra-550b-a55b(provider default reasoning) - Audited route overrideOpenRouter ·
nvidia/nemotron-3-ultra-550b-a55bforautomated_video_workflow(Task-7 exact OpenRouter endpoint pin together) - Generations5 independent runs × 9 scripts = 45 drafts, stored immutably after generation
- Run dates2026-07-29 → 2026-08-11
- Judges3-judge panel: Claude Opus 5, high effort (Anthropic) + DeepSeek V4 Flash (OpenRouter) + GPT-5.6 Sol, medium reasoning (Codex CLI): every draft scored independently by all three with the identical rubric, blind to model identity, evidence quotes required before each metric score; published score = per-metric consensus mean, verified with a Bradley-Terry refit (methodology)
- Samplingprovider-default sampling, no fine-tuning, no system-prompt changes; every current-ranked model gets the identical brief, style contract and research packet
- What stays privatethe exact rubric and any non-public reference material; source-publication timing is disclosed for exceptions (why)
Frequently asked questions
How good is Nemotron 3 Ultra 550B at writing?
On ToneBench it ranks #79 of 130 with a writing Elo of 1394 and an overall score of 76.8/100. That score comes from writing our 9 real YouTube scripts five times each and scoring every draft blind against our own finished versions across nine writing dimensions.
Is Nemotron 3 Ultra 550B the best NVIDIA model for writing?
Yes. Among the NVIDIA configs we tested, Nemotron 3 Ultra 550B is the strongest writer on the board.
Is Nemotron 3 Ultra 550B good value for the money?
It costs about $0.011/task (estimated). DeepSeek V4 Flash (native chat alias) scores higher for less, so it is not the value pick.
What are Nemotron 3 Ultra 550B's strengths and weaknesses?
Its strongest dimension is Visual Cues (66th on the board, 81.4). Its weakest is Anti-Slop (102nd on the board, 74.3). The full nine-metric breakdown is on this page.
Does Nemotron 3 Ultra 550B write some formats better than others?
Yes. Its best of our 9 scripts was the short explainer (80.4) and its weakest was the personal technical walkthrough / agentic workflow case study (71.2).
How consistent is Nemotron 3 Ultra 550B between runs?
We run every script 5 times. Nemotron 3 Ultra 550B's overall score varies by about ±4.7 points between runs, versus a board median of ±3.8. That is typical consistency for this board. The full per-metric variability table is on this page.
How was Nemotron 3 Ultra 550B evaluated?
Via OpenRouter using the exact model/route id nvidia/nemotron-3-ultra-550b-a55b, run on 2026-07-29. 5 runs per script, provider-default sampling, no fine-tuning; every draft scored blind with a fixed rubric by a three-family judge panel. Full details in the 'How we ran it' section and the methodology.
Is Nemotron 3 Ultra 550B open source?
Yes, it is an open-weights model. Among open-weights models it ranks 19th of 47 for writing.