Atlas

Models

← All models

Nemotron 3 Ultra 550B A55B

NVIDIA · 2026-06-04 · 560.5B parameters

Nemotron 3 Ultra 550B A55B is NVIDIA's open-weight, post-trained frontier reasoning model in the 550-billion-parameter class, with a publicly reported 55 billion active parameters per token, released on June 4, 2026. It uses a hybrid LatentMoE architecture combining Mamba-2, MoE, and attention layers with MTP, supports configurable reasoning, and has a documented maximum context of 1,048,576 tokens even though the checkpoint configuration defaults to 262,144. NVIDIA pre-trained it with an NVFP4 recipe and post-trained it with SFT, reinforcement learning, and MOPD for coding, math, instruction following, tool use, and long-context agentic workflows; the BF16 checkpoint is released under OpenMDW-1.1.

Benchmark scores

BenchmarkScore
A-Fantasia73.7
A-Fantasia (Valid Attempts with Retries)73.3
A-Fantasia Backwards Spelling89.0
A-Fantasia Chess59.0
A-Fantasia Chess (Valid Attempts with Retries)59.0
A-Fantasia Cube (Valid Attempts with Retries)73.0
A-Fantasia Cube Rotation73.0
A-Fantasia Spelling (Valid Attempts with Retries)88.0
AA-3D v1.0 Elo573
AA-AnalystAgent (pass@1)25.0
AA-AnalystAgent (pass@5)43.8
AA-AnalystAgent (pass^5)6.3
AA-Briefcase Elo878
AA-Briefcase v1.1 Elo877
AA-LCR67.0
AA-LCR v1.179.3
AA-Omniscience Index-0.4
AA-Omniscience v1.1 Accuracy21.2
AA-Omniscience v1.1 Hallucination Rate22.6
AA-Omniscience v1.1 Index3.4
AIME 202694.2
Artificial Analysis Agentic Index27.4
Artificial Analysis Coding Index49.3
Artificial Analysis Cyber Index v113.1
Artificial Analysis Intelligence Index37.8
Artificial Analysis Intelligence Index v4.323.4
Artificial Analysis Intelligence Index v4.3.222.9
Artificial Analysis Intelligence Index v5.011.8
Artificial Analysis Omniscience Accuracy22.6
Artificial Analysis Omniscience Hallucination Rate29.7
Loading Atlas data…