NVIDIA Nemotron 3 Ultra: Hybrid SSM/MoE Architecture Launch and Benchmarks
Synthesis history
6 versions, newest first.
-
Version 6 2026-06-09 18:47 UTC · 163 items
The one substantive addition this pass is Kilocode's June 9 report that Nemotron 3 Ultra ranks #1 among open-weight models on pinchbench, an OpenClaw-focused benchmark [^27622] — a data point that partially offsets the …
-
Version 5 2026-06-08 18:35 UTC · 157 items
SemiAnalysis extended their critique on June 7 with specific TerminalBench benchmark data showing Nemotron 3 Ultra is defeated not just by Kimi K2.6 but also by GLM5.1 on coding tasks [^26504], adding a second comparato…
-
Version 4 2026-06-07 18:14 UTC · 146 items
Sebastian Raschka's independent academic curation (June 6) is the one substantive addition this pass: he named the Nemotron 3 Super architecture paper the most practically important of the first half of 2026 and framed …
-
Version 3 2026-06-06 08:18 UTC · 130 items
The Nemotron Coalition, previously characterized only through SemiAnalysis's criticism, is now fully identified: NVIDIA confirmed eight named member labs (Mistral AI, Nous Research, H Company, Prime Intellect, Black For…
-
Version 2 2026-06-05 18:33 UTC · 72 items
SemiAnalysis emerged on June 5 as a new voice with sharp ideological criticism of the 'Nemotron Coalition' — an apparent NVIDIA-organized consortium for open-source model development not previously mentioned in the thre…
-
Version 1 2026-06-05 02:17 UTC · 65 items
NVIDIA launched Nemotron 3 Ultra, a 550B-total-parameter (55B active) hybrid Mamba SSM/MoE model, at Jensen Huang's Computex 2026 keynote and shipped it on June 4, 2026 [^24876][^24834]. The model is positioned as the m…