Revision History
Nemotron 3 · 15 revisions
Sizes are character counts of the article source. The signed number is the change from the previous revision.
Recent edit summaries
Detailed summaries recorded by editors. Generic maintenance summaries are omitted here; every recorded revision remains below. Dates describe the edit, not necessarily the event it covers.
- Correction: NemoClaw was not new in July 2026
Version 15 · Sep 28, 2026, 02:20 PM
- Correction: the NIM benchmark's "Native 256K max context" is a serving-configuration label, not the model's native maximum (Ultra supports 1M tokens, as the article's own tables say); Artificial Analysis launch-note scores are integers (48 and 54); Mamba state-cache mechanism and the 50 TPS target reworded to what NVIDIA states
Version 14 · Sep 16, 2026, 09:17 AM
- Correction: the 300 tokens/s figure for Nemotron 3 Ultra was Artificial Analysis's measurement on a DeepInfra BF16 endpoint, not an NVIDIA claim; SWE-bench cost figure is "up to 30%"; added NIM 2.0.12 serving section
Version 13 · Sep 16, 2026, 08:40 AM
- Added a subsection on derivatives and community use of Nemotron 3 Nano: Nemotron-Labs-TwoTower diffusion model, the Kaggle NVIDIA Nemotron Model Reasoning Challenge (LoRA rank 32 cap, 5,000+ participants, Meta AIRA3 8th place), the Nano base checkpoint, and Jetson deployment pointer; nine references appended
Version 12 · Sep 7, 2026, 03:23 AM
- Added a subsection on the Nemotron-3-Nano-CC and Ultra-CC competitive-programming variants and the unofficial IOI 2026 run (535.4 of 600), with three new references.
Version 11 · Sep 5, 2026, 02:59 PM