New Architecture | Neural Amp Modeler - Architecture 2 (A2)

Published: 02.06.2026

The next generation of the Neural Amp Modeler is here. Architecture 2 (A2) was developed by Tone3000 in collaboration with Steve Atkinson and unveiled on June 2, 2026.

Architecture 2 (A2) is the next generation of the Neural Amp Modeler (NAM), designed to overcome the limitations of the previous architecture (now referred to as A1). The main goal of A2 is to create a lighter, better, and more “slimmed-down” model that is specifically optimized for use on embedded systems (such as guitar pedals) without sacrificing accuracy.  The development is significantly funded by TONE3000 and is carried out in close collaboration with the industry to better meet technical requirements.

The tone3000.com website states:

"A2 marks a turning point for the industry in three ways:

  • Incredibly accurate sound. A2 beats Neural DSP V2, IK Multimedia ToneX, and Line 6 Proxy by a wide margin in both quantitative and blind listening tests (shown below).
  • Works anywhere. A2 runs in DAWs and even budget multi-fx pedals, with sound quality better than Quad Cortex at 50% CPU on a $3 ARM Cortex-M7 600MHz chip.
  • Fully open source. Any hardware or software maker can support A2 in their product with code that's freely available online.

A2 makes TONE3000 and Neural Amp Modeler possible on mass-market hardware.

Blackstar, Lava Music, Darkglass, HeadRush, Chaos Audio and Dimehead are all supporting A2, with dozens of major companies announcing support later this year."


Availability for Users

Availability depends on the exact time and the user’s role (as of June 2026):

  • For end users (players): A2 is generally available, but is still in the final implementation phase. The first versions of the core (NeuralAmpModelerCore v0.4.0 and v0.5.0) were released in early 2026 to enable processing of A2 models. However, work is still underway on optimized implementations (planned for v0.5.1) to fully leverage performance.
  • Models: TONE3000 has announced that it will retrain all A1 profiles available on its platform for A2. Once this process is complete, hundreds of thousands of sounds will be immediately available as A2 models.
  • Training: Anyone using the official Colab notebook or TONE3000’s online trainer now will receive A2 models by default.


Technical Innovations

A2 offers three major improvements over its predecessor, A1:

  • Lower resource requirements: A2 is designed to consume less CPU power. This enables native use on less powerful hardware without the need to first convert models to other formats. 
  • Slimmable NAM: This is a key innovation. A single A2 model can dynamically adjust its size and computational intensity. It runs at full quality on a powerful studio computer and automatically scales down to function on a guitar pedal with limited processing power. 
  • Improved Sound: Through a more fundamental examination of the architecture (as opposed to the intuitive development of A1), A2 promises greater realism in dynamics, gain structure, and frequency response.


Backward Compatibility

A key aspect of the introduction of A2 is full backward compatibility:

  • Existing models: All legacy A1 models (Standard, Lite, Feather, Nano) continue to work seamlessly in NeuralAmpModelerCore and the plugin. 
  • Projects: Existing DAW projects that use A1 models remain fully usable. 
  • Training: It is still possible to train A1 models if desired for specific applications. A2 replaces the standard but does not delete the old one.


Plugin Update Required

To use the new NAM A2 files, a software update is required. Previous versions of the Neural Amp Modeler plugin (based on older versions of NeuralAmpModelerCore) cannot load the new architecture.  A2 is not a simple file format update, but a fundamentally new neural network architecture that must be explicitly supported by the inference engine. 

To use it, you need at least the following versions:

  • NeuralAmpModelerCore: Version v0.5.2 or higher (this version adds A2 support). 
  • NeuralAmpModelerPlugin: Version v0.7.14 or higher. 
  • Training Code: Version v0.13.0 or higher.

Links

Videos