Benchmark Almanac

Nemotron-Cascade 2 30B-A3B

Nemotron-Cascade 2 30B-A3B ranks first on 2 of the 26 benchmarks it has scores on.

NVIDIA Released 19 Mar 2026 Family Nemotron-Cascade The lab's model card
27
Benchmarks with a score
2
Ranked first, out of five or more models
5
In the top three
NVIDIA Mar 2026

Nemotron-Cascade 2 30B-A3B

Open weights Small

No Epoch Capabilities Index score yet
  • IFBench 82.9% #1 /70
  • SciCode 36.4% #38 /47
  • AA-LCR (Artificial Analysis Long Context Reasoning) 39.1% #29 /31

Where it ranks

Its best score on each benchmark, and where that ranks among every model with a score there. We rank each model once, at its best setting.

Knowledge and language

BenchmarkScoreRankSource
IFBench 82.9% #1 of 70 Model card
RULER 99% #1 of 38 Model card
Arena-Hard v2.0 83.5% #3 of 24 Model card
MMLU-Pro 79.8% #78 of 151 Model card
WMT24++ 84.1% #8 of 12 Model card
Humanity's Last Exam 17.7% #110 of 147 Model card
MMLU-Redux 86.3% #29 of 38 Model card
MMLU-ProX 72.5% #13 of 17 Model card
MultiChallenge 45.3% #29 of 36 Model card
AA-LCR (Artificial Analysis Long Context Reasoning) 39.1% #29 of 31 Model card
LongBench v2 40.3% #26 of 27 Model card
CL-bench 12.2% #6 of 6 Model card

Math and science

BenchmarkScoreRankSource
AIME 2025 92.4% #38 of 143 Model card
HMMT 2025 94.6% #14 of 49 Model card
IMO-ProofBench 72.9% #2 of 5 Model card
GPQA Diamond 76.1% #134 of 284 Model card
IMO 2025 83.3% #3 of 6 Model card
IMO-AnswerBench 79.3% #25 of 37 Model card
AIME 2026 90.9% #27 of 34 Model card

Coding

BenchmarkScoreRankSource
LiveCodeBench v6 87.2% #14 of 76 Model card
Codeforces rating 2,320 #13 of 41 Model card
ICPC World Finals 2025 10.0 #2 of 3 Model card
SciCode 36.4% #38 of 47 Model card
SWE-bench Verified 50.2% #125 of 150 Model card

Agents and tools

BenchmarkScoreRankSource
tau2-bench 58.9% #60 of 69 Model card
BFCL V4 52.9% #22 of 24 Model card
Terminal-Bench 2.0 21.1% #60 of 61 Model card

Its launch post

The benchmarks the lab chose to headline, with the scores it gave.

Its system card

Every result in the longer documents the lab published with it. 1 of them isn't in the launch post.