NVIDIA: Nemotron 3 Nano hybrid reasoning-training resource

What changed

Trained and openly documented a hybrid Mamba-Transformer expert model with granular sparse routing and multi-environment post-training for reasoning and tool-use workloads.

Scientific scope

Trained and openly documented a hybrid Mamba-Transformer expert model with granular sparse routing and multi-environment post-training for reasoning and tool-use workloads. Hybridization builds on Jamba, Mamba2, GQA and granular MoE rather than originating those components. Throughput claims depend on generation lengths and GPU configuration. Later Super/Ultra and LatentMoE capabilities are not backdated from the launch roadmap.

Evidence

The linked original sources describe the stated contribution and its evidence limits. WRS has not independently rerun the experiments or certified a proof. Hybridization builds on Jamba, Mamba2, GQA and granular MoE rather than originating those components. Throughput claims depend on generation lengths and GPU configuration. Later Super/Ultra and LatentMoE capabilities are not backdated from the launch roadmap.

Contributors

• Source-listed authors and technical contributors · SCIENCE: Source-listed authors or explicitly credited technical contributors for the cited versions. Family entries retain source-specific attribution; a combined list is not one original-paper byline or equal individual work. Names alone do not establish resolved identities.

• Nemotron 3 Nano: Open, Efficient Mixture-of-Experts Hybrid Mamba-Transformer Model for Agentic Reasoning
• NVIDIA: Nemotron 3 Nano hybrid reasoning-training resource — official company record

Provisional WRS assessment, edition 2026-10-16.1. Independent expert calibration is pending. This grade does not certify correctness or assign ownership.

Full scientific record · Browse contribution grades · Request a correction