Nemotron 3 Super
Hacker ApprovedRanked 24 of 39 open-weight language models by recommenders
Open-weight Nvidia model that people run locally and praise for fast inference and open training data. Critics say it trails Qwen models, especially at coding.
Open-weight Nvidia model that people run locally and praise for fast inference and open training data. Critics say it trails Qwen models, especially at coding.
23 people gave an opinion
- recommend it
- 20
- criticize it
- 3
- approval share
- 87%
Recommended in 7 of 13 months
- November 2025: not recommended
- December 2025: not recommended
- January 2026: not recommended
- February 2026: not recommended
- March 2026: recommended
- April 2026: recommended
- May 2026: recommended
- June 2026: recommended
- July 2026: recommended
- August 2026: recommended
- September 2026: not recommended
- October 2026: recommended
- November 2026: not recommended (month in progress)
Why it carries the badge
- 20 people recommend itNeeds 10 or more, met.
- 6.7 recommenders for every criticNeeds 2 to 1 or better, met.
- Recommended in 7 of 13 monthsNeeds 4 or more, met.
The same three rules apply to every product, and the badge can’t be bought. How the badge works
What people say about it
The reasons people give, with the number of comments that give each one. Open a reason to read what they said.
Why people recommend it
Each mark is one comment
Fast inference and high throughput
“It offers ~same capabilities as gpt-oss-120b but with faster inference”
Comment from 12 Mar 2026 in “Nvidia Nemotron 3 Super” “Nvidia's nemotron recent 120B model also has some really crazy tps”
Comment from 19 Oct 2026 in “DeepSeek V4.1 Flash 1017 (Max)” “Nemotron-3-Super: 14070 tok/s prefill, ~194.5 tok/s generation.”
Comment from 2 Apr 2026 in “Google releases Gemma 4 open models” - Comment from 19 Oct 2026 in “Nvidia Nemotron 3.5 Super VL Powers Specialized Agents with Multimodal Reasoning”
Open training data and recipes
“I do really like that they have been transparent in releasing most of its training data.”
Comment from 13 Mar 2026 in “Can I Run AI locally?” “probably most important: training datasets and training recipe available”
Comment from 12 Mar 2026 in “NVIDIA Nemotron 3 Super” “Nvidia provides almost everything you need to replicate it from scratch”
Comment from 24 Aug 2026 in “OpenAI: GPT 5.6 Sol price reduction (until at least Nov 21)” Runs well on local hardware
“running nearly everything I do locally”
Comment from 13 Mar 2026 in “Can I Run AI locally?” “we can task our nemotron-3-super model to run over night when our desktops are idle”
Comment from 18 Mar 2026 in “Nvidia greenboost: transparently extend GPU VRAM using system RAM/NVMe” Holds up in agentic, long-context work
“this thing has an ability to hold attention through complicated (often noisy) agentic environments”
Comment from 2 Apr 2026 in “Google releases Gemma 4 open models” “Nemotron super 3 110B works well for 1M context long vibecoding sessions”
Comment from 15 Jun 2026 in “Ask HN: Has anyone replaced Claude/GPT with a local model for daily coding?”
What people warn about
Each mark is one comment
Weaker at coding than Qwen
“Nemotron Super 120b is great at a lot of stuff but it also seems to be not as good at coding as Qwen.”
Comment from 15 Jun 2026 in “Ask HN: Has anyone replaced Claude/GPT with a local model for daily coding?” “does poorly on agentic coding that I've given it”
Comment from 11 May 2026 in “Running local models on an M4 with 24GB memory” Trails newer Qwen models overall
“well behind of Qwen 3.8 flash next on pretty much everything except generation speed”
Comment from 15 Oct 2026 in “LFM2.5-24B-A2B: Built for fast on-device agents” “this 3.5 Super is still below Qwen 3.8 next as they state themsleves”
Comment from 15 Oct 2026 in “Nvidia Nemotron 3.5 Super VL 120B A12B”
Other open-weight language models people recommend
The four most recommended other open-weight language models, with the same counts for each, so you can compare like for like. The tick on each bar marks the 2 to 1 line the badge needs.
| Product | Recommend | Criticize | Approval | Months | Owned | Badge |
|---|---|---|---|---|---|---|
| Qwen 3.8 27B | 1,173 | 426 | 73% | 13 | 2 mo32 owners | Approved |
| Z.ai GLM | 1,043 | 384 | 73% | 13 | 2 mo52 owners | Approved |
| Kimi | 931 | 359 | 72% | 13 | 1 mo26 owners | Approved |
| Gemma 4 | 713 | 249 | 74% | 13 | 1 mo14 owners | Approved |
| Nemotron 3 SuperThis page | 20 | 3 | 87% | 7 | — | Approved |
What people said
29 of the 30 comments that mention it, newest first. Each links to the original comment.
“This ~5x improvement is definitely a step forward!”
“Nvidia's nemotron recent 120B model also has some really crazy tps”
“this 3.5 Super is still below Qwen 3.8 next as they state themsleves”
“well behind of Qwen 3.8 flash next on pretty much everything except generation speed”
“Nvidia provides almost everything you need to replicate it from scratch”
“I find them very useful for RAG-like problems”
“we're using Nemotron-3-super with incredible success”
“The setup is working REALLY well for us”
“Nemotron super 3 110B works well for 1M context long vibecoding sessions”
“Nemotron Super 120b is great at a lot of stuff but it also seems to be not as good at coding as Qwen.”
“this is in the range of Qwen 3.6, Gemma 4, Nemotron 3 Super”
“does poorly on agentic coding that I've given it”
“Nemotron models specifically, because of their Mamba2 hybrid SSM architecture, scale exceptionally well”
“more than sufficient for my needs”
“Nemotron 3 Super is our most successful model yet.”
“I abused this last week with the nemotron super to test out some stuff”
“Nemotron-3-Super: 14070 tok/s prefill, ~194.5 tok/s generation.”
“That really bogs down agentic tooling.”
“this thing has an ability to hold attention through complicated (often noisy) agentic environments”
“I gave it a whirl but was unenthused”
“we can task our nemotron-3-super model to run over night when our desktops are idle”
“it failed at some very basic tasks”
“it's pretty competitive for its size class”
“I do really like that they have been transparent in releasing most of its training data.”
“Currently, Nemotron 3 Super using Unsloth's UD Q4_K_XL quant is running nearly everything I do locally”
“running nearly everything I do locally”
“probably most important: training datasets and training recipe available”
“It offers ~same capabilities as gpt-oss-120b but with faster inference”
“the NVFP4 benchmark numbers also look fantastic”
Counts come from 1,795,640 Hacker News comments, November 2025 to November 2026. How we count. Report a mistake.