whatmodel.sh / Nemotron 3 Ultra 550B-A55B
Nemotron 3 Ultra 550B-A55B
↗ share
+ contribute
Nemotron 3 Ultra 550B-A55B is an open-weights language model from NVIDIA, released June 2026. Its published size is 550B parameters (55B active per token) and a 1M-token context window. Licence: openmdw.
$ how it compares
The LLMs ranked either side of Nemotron 3 Ultra 550B-A55B, out of 828 with a score, then more from NVIDIA.
| # | model | score | where | ||
|---|---|---|---|---|---|
| 96 |  Inkling (975B-A41B) | 70±1 | api $1.73/M +local | ||
| 97 |  Muse Glimmer 30B | 70±1 | api $0.64/M +local | ||
| 98 | Nemotron 3 Ultra 550B-A55B | 69±1 | api $1.35/M +local | ||
| 99 |  Qwen3.6 27B | 69±1 | api $1.20/M +local | ||
| 100 |  Ring-2.6-1T | 69±2 | comparison only | ||
| more from NVIDIA | |||||
| 155 | Nemotron Cascade 2 30B-A3B | 66±1 | local | ||
| 168 | Nemotron 3 Super 120B-A12B | 65±1 | api $0.28/M +local | ||
| 177 | Nemotron-Labs-3 Puzzle 75B-A9B | 64±1 | local | ||
$ what it's good at
agent tasks▁▁▃▆█▆▅▅▅▄▃▁ #83 of 300 · 22% of its score · 7 tests
Beats Grok 4.20 (0309 reasoning), GPT-5.1, Grok 4.3 here, though they rank higher overall. Trails Hy3 preview (295B-A21B), MiniMax-M2.7, Step 3.5 Flash (196B-A11B) here, though they rank lower overall.
| test | result | vs the others in its run | counts |
|---|---|---|---|
| SWE-bench Verified lab | 70.7% ✎ | █····██·████ beats 0/6 | 6.0% |
| SWE-bench Multilingual lab | 67.7% ✎ | ▄··▄▄···▄▄·█ beats 0/6 | 4.3% |
| Terminal-Bench 2.1 lab | 56.4% ✎ | █··▄█·▄····▄ beats 4/6 | 4.2% |
| BrowseComp lab | 44.4% ✎ | ▄·▄▄···▄··█▄ beats 1/6 | 2.4% |
| Humanity's Last Exam with tools lab | 37.4% ✎ | ▄····▄·█·▄·▄ beats 0/5 | 2.2% |
| τ³-bench lab | 70.9% ✎ | ▄····▄·█·▄·█ beats 2/6 | 2.0% |
| τ³-bench telecom (tau2-bench ≥ v1.0.1) lab | 92.9% ✎ | ▄·▄···▄·▄·█▄ beats 1/6 | 1.4% |
coding▁▁▁▃▆██▇▆▄▃▁ #84 of 458 · 20% of its score · 6 tests
Beats Grok 4.20 (0309 reasoning), GPT-5.4 mini, GPT-5.1 here, though they rank higher overall. Trails Llama 3.1 Nemotron Ultra 253B v1, GPT-5.4 nano, GLM-4.7 here, though they rank lower overall.
| test | result | vs the others in its run | counts |
|---|---|---|---|
| SWE-bench Verified lab | 70.7% ✎ | █····██·████ beats 0/6 | 6.0% |
| SWE-bench Multilingual lab | 67.7% ✎ | ▄··▄▄···▄▄·█ beats 0/6 | 4.3% |
| LiveCodeBench v6 (window unspecified) lab | 89.0% ✎ | ██···█··████ beats 3/6 | 2.8% |
| SciCode subproblems lab | 44.6% ✎ | ▃····▃··█·▃▃ beats 1/6 | 2.7% |
| LiveCodeBench v6 (Feb–May 2025) | 76.4% ✎ | ▄··█····▄▄·█ beats 2/6 | 2.2% |
| SciCode (level unspecified) | 39.9% ✎ | ▄·····▄█▄▄·▄ beats 1/6 | 2.2% |
chat & writing▁▁▁▃▅▄▅█▆▆▇▃ #81 of 393 * · 4.4% of its score · 1 test
Beats LongCat 2.0 (1.6T-A48B), Inkling Small (266B-A12B), Muse Glimmer 30B here, though they rank higher overall. Trails Gemma 4 12B, Gemma 4 26B-A4B, Mistral Medium 3.5 (128B) here, though they rank lower overall.
| test | result | vs the others in its run | counts |
|---|---|---|---|
| LMArena (text) | 1426.2 ✎ | ▁▁▁▂▃▃▃▆▆▇█▃ beats 279/379 | 4.4% |
math▁▂▃▅▆▅▄█▄▁▁▁ #108 of 398 · 6.6% of its score · 3 tests
Beats Claude Sonnet 4.6, MiniMax-M3 here, though they rank higher overall. Trails Gemini 2.5 Flash, Qwen3-VL 32B Thinking, gpt-oss-120b here, though they rank lower overall.
| test | result | vs the others in its run | counts |
|---|---|---|---|
| IMO-AnswerBench lab | 88.6% ✎ | ▄····▄·▄▄·█▄ beats 3/6 | 2.4% |
| OTIS mock AIME 2024–2025 | 86.7% ✎ | ▃▃▄▃▃▃▇▇█▃▂▅ beats 135/193 | 2.3% |
| AIME 2026 | 85.4% ✎ | ▄·▄·▄▄··█··▄ beats 0/6 | 1.9% |
web search▂▁▃▆▄█▃▃▁▁·▁ #71 of 80 * · 4.6% of its score · 2 tests
Trails GPT-4.1 nano, GPT-4o mini (2024-07-18), Mistral Small 3.1 24B (2503) here, though they rank lower overall.
| test | result | vs the others in its run | counts |
|---|---|---|---|
| BrowseComp lab | 44.4% ✎ | ▄·▄▄···▄··█▄ beats 1/6 | 2.4% |
| Humanity's Last Exam with tools lab | 37.4% ✎ | ▄····▄·█·▄·▄ beats 0/5 | 2.2% |
other tests15 not in a task · 56% of its score
| test | result | vs the others in its run | counts |
|---|---|---|---|
| GPQA Diamond +1 | 85.4% ✎ | ▁▁▁▁▂▄▆▄▅▇█▅ beats 147/208 | 14% |
| Humanity's Last Exam lab +1 | 26.7% ✎ | ▄··█▄··▄·▄·▄ beats 1/6 | 8.2% |
| GPQA (subset unspecified) lab | 87.0% ✎ | ▄█▄▄·▄·····▄ beats 2/6 | 7.6% |
| MMLU-Pro lab | 86.8% ✎ | ▄·····▄▄▄·▄█ beats 3/6 | 4.4% |
| Global-MMLU-Lite | 90.9% ✎ | ▃······▅··█▃ beats 5/6 | 4.1% |
| Mystery Game Puzzles (Epoch) | 20.0% ✎ | █··▂▃▇▄█▂▃▁▁ beats 35/74 | 2.6% |
| LongBench v2 lab | 61.9% ✎ | ▂·▁·▂▂▁▃▄▁█▄ beats 25/28 | 2.5% |
| MMLU-ProX lab | 83.0% ✎ | ▄·····▄·▄▄█▄ beats 1/6 | 2.5% |
7 more tests
| test | result | vs the others in its run | counts |
|---|---|---|---|
| IFBench (aggregation unspecified) | 81.4% ✎ | ▃······█▃▃·▃ beats 6/6 | 2.4% |
| Chess Puzzles (Epoch) | 12.0% ✎ | █▁▁▁▁▃▁▄▂▂▁▁ beats 62/135 | 2.3% |
| IFBench prompt-level (loose) lab | 81.7% ✎ | ▄▄·▄··▄▄···█ beats 5/6 | 2.1% |
| MultiChallenge lab | 63.8% ✎ | ▁··········█ beats 4/6 | 1.9% |
| CritPt lab | 3.1% ✎ | ▄····▄█···█▄ beats 2/6 | 1.0% |
| WMT25 Terminology Translation (term success rate) | 84.8% ✎ | ▂·▂··▂·▄▂·█▆ beats 4/12 | 0.7% |
| WMT26 General MT (GEMBA-ESA, LLM judge) | 67.1% ✎ | ▄·▄▄▄·▄██▄█▄ beats 1/12 | 0.3% |
$ where to run it
Through an API
api · Together AI · $1.35/M · $0.6 in / $3.6 out per 1M tokens · 1M context · docs.together.ai ↗ · price sourceStandard base token rate from models.dev. Provider-specific tiers, long-context surcharges, caching and promotions may differ.
llama.cpp · recommended
version
- Install llama.cpp · Homebrew and winget installs name the commands llama-cli and llama-server instead.macOS / Linux
curl -LsSf https://llama.app/install.sh | shWindowsirm https://llama.app/install.ps1 | iex - Run Nemotron 3 Ultra 550B-A55B · needs ~330.5 GB · from huggingface.co/unslothchat in the terminal
llama cli -hf unsloth/NVIDIA-Nemotron-3-Ultra-550B-A55B-GGUF:UD-Q4_K_XL - Or start a local server and open the address it prints (http://localhost:9931)
llama serve -hf unsloth/NVIDIA-Nemotron-3-Ultra-550B-A55B-GGUF:UD-Q4_K_XL
LM Studio
version
- Install LM Studio · Open the app once before using the lms command.
- Run Nemotron 3 Ultra 550B-A55B · needs ~330.5 GB · from huggingface.co/unsloth
- Or download it in the terminal
lms get https://huggingface.co/unsloth/NVIDIA-Nemotron-3-Ultra-550B-A55B-GGUF@UD-Q4_K_XL -y - Then run lms ls to see its name, and lms chat followed by that name
MLX (Mac)
version
- Install MLX (Mac) · Apple silicon Macs only, text only.with uv
uv tool install mlx-lmorpip install mlx-lm - Run Nemotron 3 Ultra 550B-A55B · needs ~309.3 GB, Mac only · from huggingface.co/pipenetworkchat in the terminal
mlx_lm.chat --model pipenetwork/NVIDIA-Nemotron-3-Ultra-550B-A55B-MLX-4bit - Or start an OpenAI-compatible server at http://localhost:8080/v1
mlx_lm.server --model pipenetwork/NVIDIA-Nemotron-3-Ultra-550B-A55B-MLX-4bit
Official weights and documentation: huggingface.co/nvidia ↗
$ how the score was worked out · NVIDIA — Nemotron 3 Ultra tec… 57%, A.X K2 Technical Report (SK T… 25%, Epoch AI — internal evaluatio… 13% and 2 more
Each result implies a score on its own, after adjusting for how hard its test is and who ran it. The score is their weighted average, with a small pull toward the middle for thinly tested models. Each test's weight is its counts column above; here it's added up by who ran the tests. How it works
| source | results | share of the score |
|---|---|---|
 NVIDIA — Nemotron 3 Ultra technical report · lab-reported | 18 | 57% |
| A.X K2 Technical Report (SK Telecom), arXiv 2608.30181v1 | 7 | 25% |
 Epoch AI — internal evaluations | 4 | 13% |
 LMArena leaderboard dataset | 1 | 4% |
| Cohere — North Small Translate (arXiv:2609.13916v1) | 2 | 1% |
| prior · pull toward the middle | 0% | |
| weighted average | = 69.4 |
$ improve this page
Use a short GitHub form or edit the data through a pull request. Every number needs a source; here's which ones we can use.
data updated · sources and licences · json