AI Radar

Nemotron-3 8B

Available · Language · Milestone

Nemotron-3 8B is a family of 8-billion-parameter language models from NVIDIA, introduced on 15 November 2023 with its AI foundry service on Microsoft Azure: a base model, three chat models aligned with different methods and a question-answering model. The weights were offered through Hugging Face, the NGC catalog and the Azure AI model catalog under an NVIDIA community licence. [1] [2] [3] Primary source

Timeline of Nemotron-3 8B →

Claims and evidence

  • Released Primary source[1]dateline; Curated, Optimized Models for Custom Generative AI [2]page date; introduction
  • Status AvailableWeights remain on NVIDIA's Hugging Face organisation; downloads are gated with manual approval under the NVIDIA AI Foundation Models Community License Agreement. Primary source[3]access notice; License
      • Feature MultilingualNVIDIA says the base model is proficient in 53 languages. Primary source[2]Nemotron-3-8B base [3]Dataset & Training
      • Open weights YesDownloadable from Hugging Face and the NGC catalog after accepting the NVIDIA AI Foundation Models Community License Agreement; Hugging Face access is gated with manual approval. Primary source[3]access notice [2]introduction
      • Context window 4,096 tokens tokens Primary source[3]Description
      • Parameters 8 billion Primary source[3]Description [1]Curated, Optimized Models for Custom Generative AI
      • Access Open-weights download, cloud partnerOffered through the NVIDIA NGC catalog, Hugging Face and the Azure AI model catalog. Primary source[2]introduction [1]Curated, Optimized Models for Custom Generative AI

      Lineage

      Predecessors

      No known predecessor.

      Successors

      Based on

      Not derived from another model.

      Variants and derived

      None recorded.

      Siblings

      None recorded.

      All ancestors

      None.

      All descendants

      Variants

      Nemotron-3-8B-Base-4k

      Same dates as Nemotron-3 8B.

      • Variant Inline variant in this record.Base completion model intended for customisation. Primary source[3]Description; Intended use [2]Table 1

      Also known as: Nemotron-3-8B-Base, nvidia/nemotron-3-8b-base-4k

      Nemotron-3-8B-Chat-4k-SFT

      Same dates as Nemotron-3 8B.

      • Variant Inline variant in this record.Chat model instruction-tuned with supervised fine-tuning. Primary source[4]Description [2]Table 1

      Also known as: Nemotron-3-8B-Chat-SFT, nvidia/nemotron-3-8b-chat-4k-sft

      Nemotron-3-8B-Chat-4k-RLHF

      Same dates as Nemotron-3 8B.

      • Variant Inline variant in this record.Chat model tuned with reinforcement learning from human feedback, built on the SFT model. Primary source[5]Description [2]Table 1; Nemotron-3-8B chat

      Also known as: Nemotron-3-8B-Chat-RLHF, nvidia/nemotron-3-8b-chat-4k-rlhf

      Nemotron-3-8B-Chat-4k-SteerLM

      Same dates as Nemotron-3 8B.

      • Variant Inline variant in this record.Chat model customised with NVIDIA's SteerLM method, which lets users set response attributes at inference time. Primary source[6]Description [2]Table 1

      Also known as: Nemotron-3-8B-Chat-SteerLM, Nemotron-3-8B-SteerLM, nvidia/nemotron-3-8b-chat-4k-steerlm

      Nemotron-3-8B-QA-4k

      Same dates as Nemotron-3 8B.

      • Variant Inline variant in this record.Question-and-answer model fine-tuned to give concise answers. Primary source[7]Description [2]Table 1

      Also known as: Nemotron-3-8B-QA, nvidia/nemotron-3-8b-qa-4k

      Related AI Radar coverage

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

      All model releases from NVIDIA on AI Radar →