Atlas

Models

← All models

Nemotron 3 Super 120B A12B

NVIDIA · 2026-03-11 · 123.6B parameters

Nemotron 3 Super 120B A12B is NVIDIA's open-weight, post-trained 120B-class text model for reasoning and agentic workloads, released March 11, 2026; the released BF16 checkpoint contains exactly 123,611,012,096 parameters, while NVIDIA reports only rounded active counts (12.7B including embeddings, 12.1B excluding them). It introduces LatentMoE and multi-token prediction to the hybrid Mamba-2/attention architecture, offers switchable thinking, and supports up to 1,048,576 tokens, although the repository configuration and serving examples default to 262,144. Its base was pretrained over a 25-trillion-token horizon with a mixed NVFP4/BF16/MXFP8 recipe, then underwent supervised fine-tuning and reinforcement learning, so the final checkpoint's exact cumulative training-token exposure is undisclosed.

Benchmark scores

BenchmarkScore
AA-Briefcase Elo-95.3
AA-LCR60.0
AA-Omniscience Index-42.1
AIME 202691.7
ALE-Bench1017
Apex (MathArena)7.8
Apex Shortlist57.5
APEX-Agents-AA1.8
Artificial Analysis Agentic Index8.7
Artificial Analysis Coding Index37.7
Artificial Analysis Intelligence Index25.4
Artificial Analysis Omniscience Accuracy24.0
Artificial Analysis Omniscience Hallucination Rate87.0
Artificial Analysis Openness Index83.3
ArXivMath 01/202648.9
ArXivMath 02/202630.5
ArXivMath 12/202533.8
BLXBench72.3
BLXBench Coding category90.0
BLXBench Cost category92.7
BLXBench Debugging category71.2
BLXBench Hallucination category76.1
BLXBench pass rate48.4
BLXBench Reasoning category66.5
BLXBench Refactoring category64.7
BLXBench Security category64.1
BLXBench Speed category91.1
BLXBench UI category43.2
BullshitBench v2 Leaderboard54.0
BuseyBench SVG2.7
Loading Atlas data…