whatmodel.sh / Nemotron Cascade 2 30B-A3B

Nemotron Cascade 2 30B-A3B

+ contribute

Nemotron Cascade 2 30B-A3B is an open-weights language model from NVIDIA, released March 2026. Its published size is 31.6B parameters (3B active per token) and a 256k-token context window. Licence: nvidia-open-model.

65.6±1.4#136 of 719 LLMsfrom 13 benchmark results across 2 sources

$ how it compares

The LLMs ranked either side of Nemotron Cascade 2 30B-A3B, out of 828 with a score, then more from NVIDIA.

#	model	score	where	RAM	released
153	Spark-X2.5 4B iFLYTEK (SparkLLM)	66±1	local	~3.9GB	2026-08
154	MiniMax-M2.1 MiniMax	66±1	local	~79GB	2025-12
155	Nemotron Cascade 2 30B-A3B NVIDIA	66±1	local	~20GB	2026-03
156	Laguna XS 2.1 (33B-A3B) Poolside	66±2	local	~21GB	2026-06
157	LongCat-Flash-Chat (560B-A27B) Meituan	66±1	local	~333GB	2025-08
	more from NVIDIA
98	Nemotron 3 Ultra 550B-A55B NVIDIA	69±1	api $1.35/M +local	~89GB	2026-06
168	Nemotron 3 Super 120B-A12B NVIDIA	65±1	api $0.28/M +local	~73GB	2026-03
177	Nemotron-Labs-3 Puzzle 75B-A9B NVIDIA	64±1	local	~35GB	2026-06

see all LLMs · find the ones that fit your machine

$ what it's good at

agent tasksnot ranked on this task yet · 4 tests

test	result	vs the others in its run	difficulty	counts	why
SWE-bench Verified lab	50.2% ✎	▁▁▁▁▁▁▂▄█▆▁▁ beats 41/131	hard · avg 19%	—	not counted yet: no other model in the same run to compare it with
BFCL V4 (December 2025 archive) lab	52.9% ✎	▁▁▁▂▇▆▆▆█▂▄▂ beats 78/111	hard · avg 32%	—	not counted yet: no other model in the same run to compare it with
τ²-bench aggregate lab	58.9% ✎	▁▁▁·▂▂▄▃▃▅█▂ beats 10/37	medium · avg 40%	—	not counted yet: no other model in the same run to compare it with
Terminal-Bench 2.0 lab	21.1% ✎	▂▃▅▅████▆▂▃▃ beats 1/36	very hard · avg 9.5%	—	not counted yet: no other model in the same run to compare it with

coding▁▁▁▃▆██▇▆▄▃▁ #98 of 458 · 29% of its score · 5 tests

Beats Grok 4.20 (0309 reasoning), Grok 4.3, Muse Glimmer 30B here, though they rank higher overall. Trails Llama 3.1 Nemotron Ultra 253B v1 here, though they rank lower overall.

test	result	vs the others in its run	difficulty	counts	why
LiveCodeBench Pro 2025Q2 (Medium) lab	27.6% ✎	█··▄▂▂██▄·▄▆ beats 22/22	very hard · avg 3.0%	▇▇▇▇▇▇ 16%	sharp
LiveCodeBench Pro 2025Q2 (Easy) lab	87.0% ✎	▂··▃▂▃▆▅█·▃▃ beats 20/20	medium · avg 45%	▇▇▇ 8.1%	typical
SciCode (level unspecified) lab	36.4% ✎	▁····▃▄▄▄█▆▂ beats 18/34	hard · avg 19%	▇▇ 5.4%	at its level · small run
SWE-bench Verified lab	50.2% ✎	▁▁▁▁▁▁▂▄█▆▁▁ beats 41/131	hard · avg 19%	—	not counted yet: no other model in the same run to compare it with
LiveCodeBench v6 (Aug 2024–May 2025) lab	87.2% ✎	▂▅▅▃█▃▂····▃ beats 17/17	medium · avg 45%	—	not counted yet: no other model in the same run to compare it with

chat & writing▁▁▁▃▅▄▅█▆▆▇▃ #64 of 393 * · 7.3% of its score · 1 test

Beats GPT-5.2, Qwen3.8 27B, LongCat 2.0 (1.6T-A48B) here, though they rank higher overall.

test	result	vs the others in its run	difficulty	counts	why
Arena-Hard v2 aggregate lab +1	83.5% ✎	▁▁··▄▃▁▃█▁▁▇ beats 22/22	hard · avg 28%	▇▇▇ 7.3%	sharp · at its level · small run

math▁▂▃▅▆▅▄█▄▁▁▁ #96 of 398 · 39% of its score · 4 tests

Beats Claude Sonnet 4.6, MiniMax-M3, Nemotron 3 Ultra 550B-A55B here, though they rank higher overall. Trails Gemini 2.5 Flash, GLM-4.5-Air, Granite 4.2 30B here, though they rank lower overall.

test	result	vs the others in its run	difficulty	counts	why
AIME 2025 lab +1	92.4% ✎	▁·▁▁▁▂▅▆█▇▅▁ beats 122/133	medium · avg 63%	▇▇▇▇▇ 12%	sharp · maxed out
IMO-AnswerBench lab	79.3% ✎	▁▁·▁▁▁▄▅█▄▂▃ beats 20/42	medium · avg 42%	▇▇▇▇ 11%	typical
AIME 2026 lab +1	90.9% ✎	▁▁▁▂▁▂▆█▆▂·▁ beats 24/49	medium · avg 64%	▇▇▇▇ 10%	sharp · maxed out
HMMT February 2025 lab +1	94.6% ✎	▁▁▁▃▁▅▅▃██▁▄ beats 25/31	easy · avg 69%	▇▇▇ 6.8%	small run

other tests11 not in a task · 24% of its score

test	result	vs the others in its run	difficulty	counts	why
EuroEval Life in the UK (English knowledge MC), validation split, 5-shot, accuracy	93.3% ✎	▂·▂▂·▂·▃▃·█▃ beats 11/14	easy · avg 80%	▇▇ 5.7%	maxed out
EuroEval HellaSwag (English), validation split, 5-shot generative MC, accuracy	70.8% ✎	▂··▂▂·▂▆▆█·▄ beats 5/15	medium · avg 63%	▇▇ 5.3%	typical
MultiChallenge lab	45.3% ✎	▁·▁·▁▃▆▅▃▁█▅ beats 10/29	hard · avg 27%	▇▇ 5.1%	at its level · small run
MMLU-ProX lab	72.5% ✎	▁·▁▁▁▂▁▃▂█▅▄ beats 10/35	medium · avg 54%	▇▇ 4.3%	small run
CL-bench lab	12.2% ✎	█··········▄ beats 1/2	very hard · avg 7.1%	▇ 3.4%	blunt · small run
GPQA Diamond lab	76.1% ✎	▁▁▁▁▂▄▅▅███▄ beats 177/310	hard · avg 40%	—	not counted yet: no other model in the same run to compare it with
MMLU-Pro lab	79.8% ✎	▁▁·▁▁▁▃▂▄▅█▁ beats 125/188	medium · avg 57%	—	not counted yet: no other model in the same run to compare it with
Humanity's Last Exam lab	17.7% ✎	▁▁▁▃▆▆▇██▆▁▃ beats 40/104	very hard · avg 3.8%	—	not counted yet: no other model in the same run to compare it with

3 more tests

test	result	vs the others in its run	difficulty	counts	why
MMLU-Redux lab	86.3% ✎	▁▁▃▁▂▃▄▅▇▅█▄ beats 25/63	easy · avg 77%	—	not counted yet: no other model in the same run to compare it with
LongBench v2 lab	40.3% ✎	▂·▁·▂▂▁▃▄▁█▄ beats 9/28	very hard · avg 32%	—	not counted yet: no other model in the same run to compare it with
IFBench prompt-level lab	82.9% ✎	█▄▄··▄·▄···▄ beats 6/6		—	not counted yet: no other model in the same run to compare it with

$ where to run it

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 | sh
Windows
irm https://llama.app/install.ps1 | iex
other ways to install ↗
Run Nemotron Cascade 2 30B-A3B · needs ~24.7 GB · from huggingface.co/bartowski
chat in the terminal
llama cli -hf bartowski/nvidia_Nemotron-Cascade-2-30B-A3B-GGUF:Q4_K_M
Or start a local server and open the address it prints (http://localhost:9931)
llama serve -hf bartowski/nvidia_Nemotron-Cascade-2-30B-A3B-GGUF:Q4_K_M

LM Studio

version

Install LM Studio · Open the app once before using the lms command.
download it (macOS / Windows / Linux) ↗
Run Nemotron Cascade 2 30B-A3B · needs ~24.7 GB · from huggingface.co/bartowski
open in LM Studio ↗
Or download it in the terminal
lms get https://huggingface.co/bartowski/nvidia_Nemotron-Cascade-2-30B-A3B-GGUF@Q4_K_M -y
Then run lms ls to see its name, and lms chat followed by that name

Ollama

Install Ollama
download it (macOS / Windows) ↗
Linux
curl -fsSL https://ollama.com/install.sh | sh
Run Nemotron Cascade 2 30B-A3B · needs ~19.46 GB · from ollama.com
chat in the terminal
ollama run nemotron-cascade-2:30b

MLX (Mac)

version

Install MLX (Mac) · Apple silicon Macs only, text only.
with uv
uv tool install mlx-lm
or
pip install mlx-lm
Run Nemotron Cascade 2 30B-A3B · needs ~17.8 GB, Mac only · from huggingface.co/mlx-community
chat in the terminal
mlx_lm.chat --model mlx-community/Nemotron-Cascade-2-30B-A3B-4bit
Or start an OpenAI-compatible server at http://localhost:8080/v1
mlx_lm.server --model mlx-community/Nemotron-Cascade-2-30B-A3B-4bit

Official weights and documentation: huggingface.co/nvidia ↗

$ how the score was worked out · NVIDIA — Nemotron-Cascade 2 t… 88%, EuroEval leaderboards (result… 11%

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	licence
NVIDIA — Nemotron-Cascade 2 technical report · lab-reported	11	▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇ 88%	CC BY 4.0
EuroEval leaderboards (results collection), English generative leaderboard	2	▇▇▇ 11%	MIT
prior · pull toward the middle		▇ 1%
weighted average		= 65.6

not counted (15)

nvidia/Nemotron-Cascade-2-30B-A3B model card	AIME 2025 [correct]	92.4%	the only model in this source's run of the test, so there is nothing to compare it with there; another run of the test counts
nvidia/Nemotron-Cascade-2-30B-A3B model card	AIME 2026 [correct]	90.9%	the only model in this source's run of the test, so there is nothing to compare it with there; another run of the test counts
nvidia/Nemotron-Cascade-2-30B-A3B model card	HMMT February 2025 [correct]	94.6%	the only model in this source's run of the test, so there is nothing to compare it with there; another run of the test counts
nvidia/Nemotron-Cascade-2-30B-A3B model card	LiveCodeBench v6 (Aug 2024–May 2025) [correct]	87.2%	fewer than three models have this test yet, or this source ran it for this model alone
nvidia/Nemotron-Cascade-2-30B-A3B model card	MMLU-Redux [correct]	86.3%	fewer than three models have this test yet, or this source ran it for this model alone
nvidia/Nemotron-Cascade-2-30B-A3B model card	MMLU-Pro [correct]	79.8%	fewer than three models have this test yet, or this source ran it for this model alone
nvidia/Nemotron-Cascade-2-30B-A3B model card	GPQA Diamond [correct]	76.1%	fewer than three models have this test yet, or this source ran it for this model alone
nvidia/Nemotron-Cascade-2-30B-A3B model card	Humanity's Last Exam [correct]	17.7%	fewer than three models have this test yet, or this source ran it for this model alone
nvidia/Nemotron-Cascade-2-30B-A3B model card	Arena-Hard v2 aggregate [correct]	83.5%	the only model in this source's run of the test, so there is nothing to compare it with there; another run of the test counts
nvidia/Nemotron-Cascade-2-30B-A3B model card	IFBench prompt-level [correct]	82.9%	fewer than three models have this test yet, or this source ran it for this model alone
nvidia/Nemotron-Cascade-2-30B-A3B model card	LongBench v2 [correct]	40.3%	fewer than three models have this test yet, or this source ran it for this model alone
nvidia/Nemotron-Cascade-2-30B-A3B model card	BFCL V4 (December 2025 archive) [correct]	52.9%	fewer than three models have this test yet, or this source ran it for this model alone
nvidia/Nemotron-Cascade-2-30B-A3B model card	τ²-bench aggregate [correct]	58.9%	fewer than three models have this test yet, or this source ran it for this model alone
nvidia/Nemotron-Cascade-2-30B-A3B model card	Terminal-Bench 2.0 [correct]	21.1%	fewer than three models have this test yet, or this source ran it for this model alone
nvidia/Nemotron-Cascade-2-30B-A3B model card	SWE-bench Verified [correct]	50.2%	fewer than three models have this test yet, or this source ran it for this model alone

$ published speed measurements · 1

Reference results for the hardware, engine and settings below. They aren't verified speeds of the downloads above or predictions for your machine.

AMD Ryzen AI Max+ 395 (Strix Halo), Radeon 8060S iGPU; Beelink GTR9 Pro 128GB

IQ4_XS

78.95 tok/s

llama.cpp 505b1ed15 (build_number 10034 as recorded), Vulkan (RADV) · llama-bench; tg128; 128 generated tokens; context depth 0; n_prompt=0 row; n_batch=2048; n_ubatch=512; flash_attn=1; type_k=f16; type_v=f16; n_gpu_layers=999; threads=16; one sequence

Author's own llama-bench CSV row (avg_ts mean over repetitions, stddev_ts retained). GPU string as recorded: Radeon 8060S Graphics (RADV STRIX_HALO). Unified memory; the guide reports ~124GiB OS-visible. Kernel/Mesa/power profile per the run folder notes. Not prompt processing or multi-request throughput. Curated in the guide's data/benchmarks.csv line 12 (2026-07-16, measured-local): Current-model direct scout; reasoning and forced no-think sheep checks both answered 9, but the no-think prefix did not suppress visible reasoning. Not a broad quality evaluation.

source ↗

$ improve this page

Update model information Add a place to run it Add a test result Propose a benchmark Correct a benchmark score Edit data on GitHub

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 2026-10-11 02:20 UTC · sources and licences · json