AI Radar

Nemotron-4 15B

Research only · Language · Milestone

Nemotron-4 15B is a 15-billion-parameter multilingual language model from NVIDIA, described in a technical report posted on arXiv on 26 February 2024. It was trained on 8 trillion tokens of English, multilingual and source-code text; NVIDIA says it was built as a general-purpose model that fits on a single A100 or H100 GPU. [1] Primary source

Timeline of Nemotron-4 15B →

Claims and evidence

  • Announced Primary source[1]Submission history, v1
  • ReleasedRelease date unknown
  • Status Research onlyDescribed in a technical report only. As of 2026-10-01 no NVIDIA weight release (Hugging Face or NGC) or NVIDIA-hosted access for this model was found. Primary source[1]
  • Successor of Nemotron-3 8B EditorialEditorial link along the Nemotron generation numbers: the report introduces Nemotron-4 without naming Nemotron-3 8B as its predecessor. Primary source[1]
  • Change · Size 15 billion parameters, up from 8 billion for Nemotron-3 8B.Compared with Nemotron-3 8B Primary source[1] [2]
  • Change · Training data Pre-trained on 8 trillion tokens covering 43 programming languages, against 3.8 trillion tokens and 37 programming languages for the Nemotron-3 8B base model.Compared with Nemotron-3 8B Primary source[1] [2]
  • Change · Architecture Uses grouped-query attention, rotary position embeddings and squared ReLU activations; NVIDIA's Nemotron-3 8B model cards describe a GPT-3 style network.Compared with Nemotron-3 8B Primary source[1] [2]
  • Feature MultilingualPre-training data covers 53 natural languages besides English and 43 programming languages. Primary source[1]Abstract; section 2, Data
  • Context window 4096 tokensGiven as the sequence length in the report's hyper-parameter table. Primary source[1]Table 1, sequence length
  • Parameters 15 billionThe report splits this into 3.2 billion embedding and 12.5 billion non-embedding parameters. Primary source[1]Abstract; section 2

Lineage

Predecessors

Successors

No known successor.

Based on

Not derived from another model.

Variants and derived

Siblings

None recorded.

All ancestors

All descendants

Variants

No variants recorded in this record.

Related AI Radar coverage

AI Radar coverage starts in June 2026; no coverage linked yet.

All model releases from NVIDIA on AI Radar →