[whatmodel.sh](../../README.md) / Nemotron Cascade 2 30B-A3B # Nemotron Cascade 2 30B-A3B NVIDIA · released 2026-03 · 31.6B params (3B active) · nvidia-open-model **score 64.6** ±1.4 · #148 of 732 · from 13 benchmark results across 2 sources Score within the [LMArena (text)](https://whatmodel.sh/?set=lmarena-text&score=set) comparison set only. ## What it's good at _{● marks Nemotron Cascade 2 30B-A3B among every model on that task or test, worst to best.}

task            vs the others            rank  counts  tests
agent tasks                    not ranked yet       —      4
coding          ▁▁▁▃▅▇█▇●▄▃▁      #106 of 469     28%      5
chat & writing  ▁▁▁▃▆▄▆█▆▇●▃      #63 of 402*    6.9%      1
math            ▁▁▃▅█▆▄●▄▂▁▁      #107 of 431     40%      4

agent tasks · 0.0% of its score

test                                result  vs the others in its run   difficulty           counts  why
SWE-bench Verified (maker)           50.2%  ▁▁▁▁▁▁▂●█▆▁▁ beats 42/138  hard · avg 19%            —  not counted yet: no other model in the …
BFCL V4 (December 2025 archive) (…   52.9%  ▁▁▁▂▆▆▅▆●▂▄▃ beats 78/114  hard · avg 31%            —  not counted yet: no other model in the …
τ²-bench aggregate (maker)           58.9%  ▁▁▁·▁▁●▃▃▄█▁ beats 10/39   medium · avg 45%          —  not counted yet: no other model in the …
Terminal-Bench 2.0 (maker)           21.1%  ▂●▅▅████▆▂▃▃ beats 1/36    very hard · avg 10%       —  not counted yet: no other model in the …

coding · #106 of 469 · 28% of its score

Beats [Grok 4.20 (0309 reasoning)](grok-4-20.md), [Grok 4.3](grok-4-3.md), [Muse Glimmer 30B](muse-glimmer-30b.md) here, though they rank higher overall. Trails [Llama 3.1 Nemotron Ultra 253B v1](llama-3-1-nemotron-ultra-253b.md) here, though they rank lower overall.

test                                result  vs the others in its run   difficulty            counts  why
LiveCodeBench Pro 2025Q2 (Medium)…   27.6%  █··▄▂▂██▄·▄● beats 22/22   very hard · avg 3.5%     14%  sharp
LiveCodeBench Pro 2025Q2 (Easy) (…   87.0%  ▂··▃▂▃▆▅█·▃● beats 20/20   medium · avg 47%        7.6%  typical
SciCode (level unspecified) (make…   36.4%  ▁····▃▃▄▅●▆▂ beats 21/40   hard · avg 19%          5.8%  at its level · small run
SWE-bench Verified (maker)           50.2%  ▁▁▁▁▁▁▂●█▆▁▁ beats 42/138  hard · avg 19%             —  not counted yet: no other model in the …
LiveCodeBench v6 (Aug 2024–May 20…   87.2%  ▂▅▅▃█▃▂····● beats 17/17   medium · avg 45%           —  not counted yet: no other model in the …

chat & writing · #63 of 402 · 6.9% of its score

Beats [Claude Haiku 5.5](claude-haiku-5-5.md), [GPT-6 Luna](gpt-6-luna.md), [Grok 4.7](grok-4-7.md) here, though they rank higher overall.

test                             result  vs the others in its run  difficulty      counts  why
Arena-Hard v2 aggregate (maker)   83.5%  ▁▁··▄▃▁▃█▁▁● beats 22/22  hard · avg 30%    6.9%  sharp · at its level · small run

math · #107 of 431 · 40% of its score

Beats [Claude Sonnet 4.6](claude-sonnet-4-6.md), [MiniMax-M3](minimax-m3.md), [Nemotron 3 Ultra 550B-A55B](nemotron-3-ultra-550b-a55b.md) here, though they rank higher overall. Trails [Gemini 2.5 Flash](gemini-2-5-flash.md), [GLM-4.5-Air](glm-4-5-air.md), [Granite 4.2 30B](granite-4-2-30b.md) here, though they rank lower overall.

test                        result  vs the others in its run    difficulty        counts  why
AIME 2025 (maker)            92.4%  ▁·▁▁▁▂▅▆█▇●▁ beats 122/133  medium · avg 64%     13%  sharp · maxed out
IMO-AnswerBench (maker)      79.3%  ▁▁·▁▁▁▄▆●▄▂▃ beats 22/44    medium · avg 43%     11%  typical
AIME 2026 (maker)            90.9%  ▁▁▁▂▂▂▆●▆▂·▁ beats 25/52    easy · avg 66%       10%  sharp · maxed out
HMMT February 2025 (maker)   94.6%  ▁▁▁▃▁▅▅▃█●▁▄ beats 25/31    easy · avg 72%      6.3%  small run

other tests (11)

test                                result  vs the others in its run    difficulty            counts  why
EuroEval Life in the UK (English …   93.3%  ▂·▂▂·▂·▃▃·●▃ beats 11/14    easy · avg 80%          5.9%  maxed out
EuroEval HellaSwag (English), val…   70.8%  ▂··▂▂·▂●▆█·▄ beats 5/15     medium · avg 63%        5.6%  typical
MultiChallenge (maker)               45.3%  ▁·▁·▁▃●▅▃▂█▅ beats 10/30    hard · avg 28%          5.1%  at its level · small run
MMLU-ProX (maker)                    72.5%  ▁·▁▁▁▂▁●▃█▅▄ beats 10/36    medium · avg 54%        4.3%  small run
CL-bench (maker)                     12.2%  ●··········▄ beats 1/2      very hard · avg 7.2%    3.4%  blunt · small run
GPQA Diamond (maker)                 76.1%  ▁▁▁▁▂▄▅▅●██▄ beats 182/322  hard · avg 39%             —  not counted yet: no other model in the …
MMLU-Pro (maker)                     79.8%  ▁▁·▁▁▁▃▂▄▅●▁ beats 125/188  medium · avg 54%           —  not counted yet: no other model in the …
Humanity's Last Exam (maker)         17.7%  ▁▁▁▄▆▆●██▆▁▃ beats 43/111   very hard · avg 3.8%       —  not counted yet: no other model in the …
MMLU-Redux (maker)                   86.3%  ▁▁▃▁▂▃▄●▇▅█▄ beats 25/63    easy · avg 76%             —  not counted yet: no other model in the …
LongBench v2 (maker)                 40.3%  ▂·▁·▂▃▁●▄▃█▄ beats 10/33    very hard · avg 32%        —  not counted yet: no other model in the …
IFBench prompt-level (maker)         82.9%  █▄▄··▄·▄···● beats 6/6                                 —  not counted yet: no other model in the …

## How the score was worked out Each result implies a score on its own, after adjusting for how hard that test is and who ran it. The model's 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 is added up by who ran the tests. [Method](../method.md)

source                               results  share of the score
NVIDIA (submitted / maker-reported)       11   88% ▇▇▇▇▇▇▇▇▇▇▇
EuroEval leaderboards                      2   11% ▇
prior: pull toward the middle                   1% ▇
weighted average                              = 64.6

_{Licences: NVIDIA ([CC BY 4.0](../../licenses/report-2603.19220v2.txt)) · EuroEval leaderboards ([MIT](https://github.com/EuroEval/leaderboards/blob/1a77f1d2f8ba1ba05ab73287fb090d76bf41b2e0/LICENSE))}

Not counted (15)

source                  test                            score  why
nvidia/Nemotron-Casca…  AIME 2025                       92.4%  only model in this source's run
nvidia/Nemotron-Casca…  AIME 2026                       90.9%  only model in this source's run
nvidia/Nemotron-Casca…  HMMT February 2025              94.6%  only model in this source's run
nvidia/Nemotron-Casca…  LiveCodeBench v6 (Aug 2024–Ma…  87.2%  too few models have it yet
nvidia/Nemotron-Casca…  MMLU-Redux                      86.3%  too few models have it yet
nvidia/Nemotron-Casca…  MMLU-Pro                        79.8%  too few models have it yet
nvidia/Nemotron-Casca…  GPQA Diamond                    76.1%  too few models have it yet
nvidia/Nemotron-Casca…  Humanity's Last Exam            17.7%  too few models have it yet
nvidia/Nemotron-Casca…  Arena-Hard v2 aggregate         83.5%  only model in this source's run
nvidia/Nemotron-Casca…  IFBench prompt-level            82.9%  too few models have it yet
nvidia/Nemotron-Casca…  LongBench v2                    40.3%  too few models have it yet
nvidia/Nemotron-Casca…  BFCL V4 (December 2025 archiv…  52.9%  too few models have it yet
nvidia/Nemotron-Casca…  τ²-bench aggregate              58.9%  too few models have it yet
nvidia/Nemotron-Casca…  Terminal-Bench 2.0              21.1%  too few models have it yet
nvidia/Nemotron-Casca…  SWE-bench Verified              50.2%  too few models have it yet

## How to run it ### llama.cpp (recommended) 1. Install llama.cpp. Homebrew and winget installs name the commands llama-cli and llama-server instead. [Other ways to install](https://github.com/ggml-org/llama.cpp/blob/master/docs/install.md) ```sh # macOS / Linux curl -LsSf https://llama.app/install.sh | sh # Windows irm https://llama.app/install.ps1 | iex ``` 2. Run Nemotron Cascade 2 30B-A3B (Q4_K_M, needs ~24.7 GB, from [huggingface.co/bartowski](https://huggingface.co/bartowski/nvidia_Nemotron-Cascade-2-30B-A3B-GGUF/tree/931b595fc71b7ca14fb9d935af011f69f7c0434c)) ```sh llama cli -hf bartowski/nvidia_Nemotron-Cascade-2-30B-A3B-GGUF:Q4_K_M ``` 3. Or start a local server and open the address it prints (http://localhost:9931) ```sh llama serve -hf bartowski/nvidia_Nemotron-Cascade-2-30B-A3B-GGUF:Q4_K_M ``` Other versions: Q8_0 (~33.6 GB) `llama cli -hf bartowski/nvidia_Nemotron-Cascade-2-30B-A3B-GGUF:Q8_0` ### LM Studio 1. Install LM Studio. Open the app once before using the lms command. [Download it (macOS / Windows / Linux)](https://lmstudio.ai/download) 2. Run Nemotron Cascade 2 30B-A3B (Q4_K_M, needs ~24.7 GB, from [huggingface.co/bartowski](https://huggingface.co/bartowski/nvidia_Nemotron-Cascade-2-30B-A3B-GGUF/tree/931b595fc71b7ca14fb9d935af011f69f7c0434c)) ```sh lms get https://huggingface.co/bartowski/nvidia_Nemotron-Cascade-2-30B-A3B-GGUF@Q4_K_M -y ``` 3. Then run lms ls to see its name, and lms chat followed by that name Other versions: Q8_0 (~33.6 GB) `lms get https://huggingface.co/bartowski/nvidia_Nemotron-Cascade-2-30B-A3B-GGUF@Q8_0 -y` · 4bit (~17.8 GB, Mac only) `lms get https://huggingface.co/mlx-community/Nemotron-Cascade-2-30B-A3B-4bit -y` · 8bit (~33.6 GB, Mac only) `lms get https://huggingface.co/mlx-community/Nemotron-Cascade-2-30B-A3B-8bit -y` ### Ollama 1. Install Ollama. [Download it (macOS / Windows)](https://ollama.com/download) ```sh # Linux curl -fsSL https://ollama.com/install.sh | sh ``` 2. Run Nemotron Cascade 2 30B-A3B (Q4_K_M, needs ~19.46 GB, from [ollama.com](https://ollama.com/library/nemotron-cascade-2/tags)) ```sh ollama run nemotron-cascade-2:30b ``` ### MLX (Mac) 1. Install MLX (Mac). Apple silicon Macs only, text only. ```sh # with uv uv tool install mlx-lm # or pip install mlx-lm ``` 2. Run Nemotron Cascade 2 30B-A3B (4bit, needs ~17.8 GB, Mac only, from [huggingface.co/mlx-community](https://huggingface.co/mlx-community/Nemotron-Cascade-2-30B-A3B-4bit/tree/cf40185ae9f11417eef7f9af6f6d5e11740e52a6)) ```sh mlx_lm.chat --model mlx-community/Nemotron-Cascade-2-30B-A3B-4bit ``` 3. Or start an OpenAI-compatible server at http://localhost:8080/v1 ```sh mlx_lm.server --model mlx-community/Nemotron-Cascade-2-30B-A3B-4bit ``` Other versions: 8bit (~33.6 GB, Mac only) `mlx_lm.chat --model mlx-community/Nemotron-Cascade-2-30B-A3B-8bit` Model repository and documentation: [huggingface.co/nvidia](https://huggingface.co/nvidia/Nemotron-Cascade-2-30B-A3B) ## Published speed measurements Reference results for the hardware, engine and settings below. These are not verified speeds of the downloads above or predictions for your machine.

hardware                      precision         speed  engine and settings
AMD Ryzen AI Max+ 395 (Stri…  IQ4_XS      78.95 tok/s  llama.cpp 505b1ed15 (build_number…
Apple M5 Max (40-core GPU),…  q4_k_m     100.19 tok/s  ollama (version not recorded) · l…

_{Each speed links to its source, which has the full settings.} [interactive version](https://whatmodel.sh/model/nemotron-cascade-2-30b-a3b) · [JSON](https://whatmodel.sh/api?format=json&all=1)