nuottroisaoduoc/NVIDIA-Nemotron-3-Nano-30B-A3B-BF16-heretic

TEXT GENERATIONPricing:Input $0.2 / Output $0.8Concurrent Unit Cost:2Model Size:30BQuant:FP8Context Size:32kPublished:Aug 13, 2026License:otherArchitecture:Transformer Featherless Exclusive Cold

This is a 30 billion parameter decensored version of NVIDIA's Nemotron-3-Nano-30B-A3B-BF16, created using Heretic v1.2.0 with the Arbitrary-Rank Ablation (ARA) method. The original model, developed by NVIDIA, is a hybrid Mixture-of-Experts (MoE) architecture with 3.5B active parameters, designed for both reasoning and non-reasoning tasks with a configurable reasoning trace. This modified version significantly reduces refusals compared to the original, making it suitable for general-purpose reasoning and chat applications in English and 5 other languages, with a maximum context length of 1M tokens.

Loading preview...

Model Overview

This model is a decensored variant of NVIDIA's Nemotron-3-Nano-30B-A3B-BF16, processed using the Heretic tool with Arbitrary-Rank Ablation (ARA) to reduce content refusals. The original Nemotron-3-Nano-30B-A3B-BF16 is a 30 billion parameter large language model (LLM) developed by NVIDIA, featuring a hybrid Mixture-of-Experts (MoE) architecture with 3.5 billion active parameters. It combines Mamba-2 and Attention layers, and is designed to excel in both reasoning and non-reasoning tasks by optionally generating a reasoning trace before its final response.

Key Differentiators

  • Decensored Version: Significantly reduces content refusals (6/100 compared to 99/100 for the original model), offering broader utility.
  • Hybrid MoE Architecture: Employs a unique blend of 23 Mamba-2 and MoE layers, alongside 6 Attention layers, with 128 experts plus 1 shared expert per MoE layer.
  • Configurable Reasoning: Can generate detailed reasoning traces for complex problems, improving accuracy, or provide direct answers for faster responses.
  • Extensive Context Window: Supports a maximum input and output size of 1 million tokens, ideal for long-context tasks.
  • Multilingual Support: Supports English, German, Spanish, French, Italian, and Japanese, along with 43 programming languages.

Use Cases

This model is well-suited for developers building AI Agent systems, chatbots, RAG systems, and other AI-powered applications requiring general-purpose reasoning and instruction-following capabilities, particularly where reduced content moderation is desired.