Nvidia Nemotron 3.5 Transcribe
Comments on “Nvidia Nemotron 3.5 Transcribe”
Claim-checked. Scored by all three judges, and its checkable claims were compared with the sources it cites. How far stories are checked
Nvidia released Nemotron 3.5 Transcribe, an open-weight offline model (34B total, about 4.3B active parameters) covering transcription in 40 languages, translation, speaker labels and audio Q&A for recordings up to an hour. Its own tables show a 4.71 mean WER on nine English Open ASR sets, against 4.94 for Qwen3-ASR-1.7B. The 2,238x throughput figure comes from Nvidia's NIM on one Blackwell GPU; the open-source vLLM path is about two times slower. These results are Nvidia's own and not independently verified.
Why it ranked
No. 18 of 110 with a score of 0.66. The panel rated it highest on new information (0.94) and technical depth (0.88), lowest on audience breadth (0.57). Placed No. 18, where the score alone would also have put it: the fifth AI & machine learning story of the day (−0.040). The placement terms move stories; they are not part of the score.
- AI judgeJudge 2claude-haiku-5-5
- Major open-weight speech model release from Nvidia with benchmarks and technical architecture details; vendor-authored, benchmark claims largely self-reported, relevant to AI builders but niche to speech.
- AI judgeJudge 3gpt-5.6-luna
- A major open-weight multilingual speech-model release with extensive benchmarks, architecture details, deployment instructions, hardware support, and training-data disclosure. It is highly actionable, though some comparative claims depend on published benchmark scope.
Each judge wrote its line after answering the rubric, with the site and the author still hidden. Every answer behind this is in the table under Score.
Score
0.66total under rubric v2.3
- Significance0.65
- Reader interest0.68
- Audience breadth0.57
- Practical relevance0.80
- Substantiated0.84
- Technical depth0.88
- New information0.94
- Counts against: Promotional0.17
- Counts against: Overclaim0.24
Each judge’s answer per question (0–1), their mean and spread, and what the question adds to the score (weight × mean).
| Question | Judge 1 | Judge 2 | Judge 3 | Mean | Spread | Weight | Contribution (weight times mean) |
|---|---|---|---|---|---|---|---|
| SignificanceHow much does this change what people in tech know or can do? | 0.71 | 0.45 | 0.78 | 0.65 | 0.33 | +0.24 | +0.155 |
| Reader interestWould a well-informed senior engineer or tech founder want to read this today, and likely pass it on? | 0.60 | 0.55 | 0.90 | 0.68 | 0.35 | +0.26 | +0.178 |
| Audience breadthHow many people who build technology does it concern? | 0.48 | 0.50 | 0.72 | 0.57 | 0.24 | +0.13 | +0.074 |
| Practical relevanceHow useful is it to someone who builds software or tech products? | 0.92 | 0.60 | 0.88 | 0.80 | 0.32 | +0.22 | +0.176 |
| SubstantiatedHow well is it backed? Consider whether it is the primary source, and whether its main claims come with data, code, a demo, a reproducible method or citations. | 0.97 | 0.60 | 0.96 | 0.84 | 0.37 | +0.05 | +0.042 |
| Technical depthHow technically substantive is it? | 0.99 | 0.70 | 0.94 | 0.88 | 0.29 | +0.05 | +0.044 |
| New informationDoes it contain information absent from the stories covered in the last 7 days (list provided)? | 0.94 | 0.90 | 0.98 | 0.94 | 0.08 | +0.05 | +0.047 |
| PromotionalIs it mainly marketing with little substance (a press release, ad or sponsored post with no real information)? A launch post that carries real news is not promotional. | 0.10 | 0.30 | 0.12 | 0.17 | 0.20 | −0.19 | −0.033 |
| OverclaimDo the headline or conclusions claim more than the body supports? | 0.18 | 0.35 | 0.18 | 0.24 | 0.17 | −0.11 | −0.026 |
| Sum of contributions | +0.657 | ||||||
| Claim-check penalty | 0.000 | ||||||
| Total, as published | 0.657 | ||||||
Spread is the highest minus the lowest judge answer. marks a spread above 0.40: the judges disagree, and the story is marked contested.
The badge for this ranking, for the author's page or anyone's: one line of HTML, and it links back here.
<a href="https://rubric.hrih.si/editions/2026-10-29/18/"><img src="https://rubric.hrih.si/badge/2026-10-29/18.svg" alt="Ranked #18 on Rubric, 29 Oct 2026" width="245" height="28"></a>Claim check
Claim check: 8 no source given. Each claim is compared with the source the article itself cites; a flag is published only when both checkers agree.
No source givenBenchmark result
Nemotron 3.5 Transcribe achieves a mean WER of 4.71 on the nine-dataset Open ASR Leaderboard, ahead of the next best open-weight model Qwen3-ASR-1.7B at 4.94.
The article gives no link or citation for it.
The article says “English ASR Open ASR Leaderboard, 9 public sets Mean WER ↓ 4.71 Qwen3-ASR-1.7B 4.94”
The source says No source to compare with.
No source givenBenchmark result
Nemotron 3.5 Transcribe scores a mean cpWER of 15.80 on multi-speaker ASR, versus 16.83 for the next best published result, DiCoW v3.3 + DiariZen.
The article gives no link or citation for it.
The article says “Mean cpWER ↓ 15.80 DiCoW v3.3 + DiariZen 16.83”
The source says No source to compare with.
No source givenBenchmark result
Nemotron 3.5 Transcribe gets a 10.98 mean WER on the long-form Open ASR Leaderboard, versus 13.37 for Parakeet-TDT-0.6B-v3.
The article gives no link or citation for it.
The article says “Open ASR Leaderboard, long-form Mean WER ↓ 10.98 Parakeet-TDT-0.6B-v3 13.37”
The source says No source to compare with.
No source givenBenchmark result
Nemotron 3.5 Transcribe gets a 4.62 mean WER on the Multilingual Open ASR Leaderboard (7 languages), versus 6.49 for Qwen3-ASR-1.7B.
The article gives no link or citation for it.
The article says “Multilingual Open ASR Leaderboard, 7 languages Mean WER ↓ 4.62 Qwen3-ASR-1.7B 6.49”
The source says No source to compare with.
No source givenNumber
With NVFP4 and DFlash2 speculative decoding in the Nemotron Transcribe NIM, one RTX PRO 6000 Blackwell Server Edition GPU transcribes up to 2,238 seconds of audio per second and returns a 30-second recording's transcript in 0.13 seconds.
The article gives no link or citation for it.
The article says “one RTX PRO 6000 Blackwell Server Edition GPU transcribes up to 2,238 seconds of audio per second (NVFP4 with DFlash2 speculative decoding) and returns the transcript of a 30-second recording…”
The source says No source to compare with.
No source givenNumber
The model has 34B total parameters, of which about 4.3B are active including the audio encoder.
The article gives no link or citation for it.
The article says “Total Parameters 34B (4.3B active), including the audio encoder”
The source says No source to compare with.
No source givenNumber
The training data comprises approximately 1.72 million unique audio hours, or 6.53 million hours including repeated exposure.
The article gives no link or citation for it.
The article says “approximately 1.72 million hours across unique indexed audio examples; 6.53 million hours including repeated exposure”
The source says No source to compare with.
No source givenDate
The model was released on Build.Nvidia.com, GitHub, Hugging Face and NGC on 10/28/2026.
The article gives no link or citation for it.
The article says “Build.Nvidia.com 10/28/2026”
The source says No source to compare with.
Also covered by
Elsewhere
Elsewhere: HN 1 pts, not on front page
Recorded for context only, checked 29 Oct 2026, 05:47 UTC. It played no part in the ranking.