For about twenty years, the argument about digital amp tone went the same way. The companies with the biggest research budgets built the most convincing amp models, you paid a premium for their hardware and their proprietary formats, and if you wanted their sound you bought their box. Then a piece of free, open-source software written by one engineer quietly rearranged the whole thing.

That software is Neural Amp Modeler, or NAM. In 2026 its new architecture beat several of the biggest commercial modellers in the largest blind listening test the guitar world has run, and it costs nothing because it is open source. Within months, manufacturers scrambled to support it. But here is the problem, and the reason this article exists: "NAM compatible" on a spec sheet can mean four completely different things, and most pedals will not tell you which one you are getting. This guide explains what NAM is, why it matters, how it differs from profiling and modelling, and which pedals genuinely run it properly.

Neural Amp Modeler is an open-source amp modelling technology created by engineer Steven Atkinson, released under the MIT licence, which means it is genuinely free to use — no fees, no licensing costs, for players and manufacturers alike.

Rather than trying to recreate an amplifier by simulating its circuit, NAM uses deep learning. You feed a reference signal through a real amp or pedal, record what comes out, and the software trains a neural network to reproduce that relationship — the exact way that specific piece of gear responded to that exact input. The result is a small .nam file that plays back like the real thing.

In June 2026, TONE3000 — working in partnership with Atkinson — released NAM Architecture 2 (A2), the next generation of the technology. Then they tested it properly.

Using the MUSHRA methodology (the audio industry's gold standard for perceived sound quality, used by organisations such as the BBC and EBU), TONE3000 ran a blind listening test with over 1,000 participants across 37 unique tones, comparing recordings of real gear against anonymised digital models from TONE3000 and the leading commercial modellers. A2 came out on top, beating Neural DSP's V1 and V2 engines, IK Multimedia's ToneX and Line 6's Proxy.

Free, open source, and rated closer to the real amp than the paid alternatives. That is why every manufacturer suddenly supports it.

The technology story is interesting. The practical reasons are better.

You get amps that no modeller will ever include. Commercial modellers ship the amps that sell: Marshalls, Fenders, Mesas, 5150s. NAM's library is built by the community, which means it contains the strange and the rare — obscure boutique heads, one-off modded amps, signature rigs, vintage oddities, and the genuinely silly (people have captured tiny practice amps, car speakers and worse, and some of those sound fantastic on a record). If a piece of gear exists, somebody has probably captured it.

There is an enormous library, free. TONE3000 hosts a community library of hundreds of thousands of captures — the platform's manufacturer integrations reference a library of more than 350,000 tones — and downloading them costs nothing.

You can capture your own rig. This is the part that should excite anyone with gear they love. Capturing is free. That boutique amp, that specific pedal-into-amp combination you spent years assembling, your favourite studio chain — you can capture it, load it into a pedal, and have that exact sound available anywhere, night after night, without carrying the rig. Your tone stops depending on the venue's backline.

And it is captured from the actual source. This is the honest appeal. You are not hearing an engineer's interpretation of what a 5150 sounds like. You are hearing a measurement of a specific amp, on a specific day, with specific settings. For players who find some digital modelling a bit sterile, that distinction matters.

These three terms get used interchangeably, and they are not the same thing.

Traditional (component) modelling is the older approach, used by Line 6, Fractal Audio and others for years. Engineers study a real amplifier's circuit and build a mathematical simulation of its components and behaviour. The great advantage is that the model is interactive — you can turn the virtual gain, bass, mid and treble knobs and it responds across the whole range, like a real amp, because the circuit behaviour is being simulated.

Profiling or capturing — Kemper's profiling, IK Multimedia's ToneX, Neural DSP's Quad Cortex captures, Mooer's and Headrush's own capture systems — takes a different route. Instead of modelling the circuit, it measures how the amp responds to a test signal at one specific setting and reproduces that. It is a snapshot, and it can be extremely accurate. The trade-off is that it captures the amp as it was set. Want a different gain setting? That is a different capture.

NAM belongs to this second family — it is a neural capture technology — with three differences that matter: it is open source and free, it is not tied to any manufacturer's hardware, and in the A2 blind test it scored higher than the commercial capture systems. A Kemper profile only works on Kemper gear. A ToneX capture works in ToneX. A .nam file works anywhere that supports NAM, and increasingly that means everywhere.

The genuine limitation to acknowledge: a static capture, NAM included, is a photograph of one setting. Traditional modelling still wins on knob interactivity. (Hold that thought — it comes back in the "what's next" section.)

The process is simple enough that anyone with an interface can do it.

You play a specially designed reference signal through the amp or pedal you want to capture — a sweep or test tone containing the information the trainer needs. You record the output. You upload the input and output recordings to a trainer (TONE3000's platform, or Atkinson's own training notebook), and the neural network learns the relationship between the two.

Out comes a .nam file. That is your amp, at those settings, in a file of a few hundred kilobytes or less.

Note what is not required: no proprietary hardware, no subscription, no permission. That accessibility is why the library grew so fast.

For its first few years NAM had one serious practical problem: it was hungry.

Running a neural network in real time with low latency takes processing power. On a desktop computer that is fine — NAM ran beautifully as a plugin in a DAW, and home recordists loved it. But taking it to a gig meant taking a laptop, an interface and an audio chain to a stage, which most players sensibly refused to do. Guitar pedals run on modest embedded chips, not desktop CPUs, and early NAM models simply demanded more than those chips could deliver.

A2 changed the economics. The new architecture is not just more accurate, it is more efficient. A2-Full delivers better quality than the old A1 architecture while using around 30–40% less CPU. And the smaller A2-Lite variant was designed specifically for constrained hardware — it runs at roughly 50% CPU on an ARM Cortex-M7, the same class of chip found in many popular modelling pedals.

That is the whole reason this article can exist. A technology that used to need a laptop now fits on a pedal you can stomp on. TONE3000 have pointed out it can run on chips costing a few dollars.

When a manufacturer says "NAM support," they could mean any of these. Here is the difference.

A1 (legacy architecture) is the original NAM. It sounds good and there are a huge number of A1 captures out there. It has simply been superseded, and it is less CPU-efficient than A2.

A2-Full is the maximum-accuracy version, intended for DAWs and hardware with real processing headroom. It is the best-sounding option available, and this is the version that topped the blind listening test.

A2-Lite is the same model run at a smaller size for processing-constrained devices such as multi-effects pedals. Here is the important part: A2-Lite still outperformed every commercial modeller in the test. It is a compromise in the sense that it is smaller, not in the sense that it is bad.

One elegant detail: an A2 download is a single file that can be run as either Full or Lite. An 8-channel network (Full) and a 3-channel network (Lite) are packed into the same download, so you do not need separate versions for your laptop and your pedal.

When a manufacturer says "NAM support," they could mean any of these. Here is the difference.

A1 (legacy architecture) is the original NAM. It sounds good and there are a huge number of A1 captures out there. It has simply been superseded, and it is less CPU-efficient than A2.

A2-Full is the maximum-accuracy version, intended for DAWs and hardware with real processing headroom. It is the best-sounding option available, and this is the version that topped the blind listening test.

A2-Lite is the same model run at a smaller size for processing-constrained devices such as multi-effects pedals. Here is the important part: A2-Lite still outperformed every commercial modeller in the test. It is a compromise in the sense that it is smaller, not in the sense that it is bad.

One elegant detail: an A2 download is a single file that can be run as either Full or Lite. An 8-channel network (Full) and a 3-channel network (Lite) are packed into the same download, so you do not need separate versions for your laptop and your pedal.

This is the most important section in the article, and the thing most spec sheets obscure.

Some pedals run NAM files natively — the .nam file loads and plays as captured. Others convert your NAM file into the manufacturer's own proprietary format first, then play that. Both get advertised as "NAM support."

They are not equivalent. TONE3000's own assessment of converted models is blunt: they are "not real NAM models, and they don't have the same sound quality or accuracy." You are hearing a copy of a copy.

So when you evaluate a pedal, the question is not "does it support NAM?" It is:

1. Native or converted?

2. If native, which architecture — A2-Full, A2-Lite or A1?

3. And separately: can the hardware actually gig? (Balanced outputs, MIDI, an FX loop, enough footswitches.)

That is exactly how we ranked the units below.

Our tiers, based on firmware release notes verified in early August 2026:

S Tier — Native A2-Full

Darkglass Anagram. The most capable unit here. It runs three NAM instances simultaneously — Neural Amp, Neural Pedal and Neural Loader blocks, each independently set to Full or Lite — on a hexacore DSP, with XLR outputs and a full effects loop. KosmOS 1.16 made it the first bass multi-effects unit with native A2-Full support. Nothing else on this list has this much headroom.

HeadRush Prime and HeadRush Core. Firmware 5.1 brought genuine native support for A2-Full, A2-Lite and all four A1 sizes, with no lossy conversion. The standout convenience feature: you can browse the TONE3000 library directly from the touchscreen over Wi-Fi — no laptop, no file transfers. One limitation to know: one NAM instance per rig.

A Tier — Native A2-Lite

Mooer GE1000 / GE1000 Li. Firmware V3.1.0 removed the converter; .nam files now load and run directly, in Mooer's own words "without format conversion." Balanced XLR outputs, switchable serial/parallel loop, 192kHz USB audio and 5-pin MIDI make it a serious gigging box that quietly became a serious NAM box. (Note: the model is the GE1000, not "GE1000 Pro.")

Mooer GS1000 / GS1000 Li. Native A2-Lite as of V1.3.0. A1 files still route through the converter, and Mooer are commendably clear about that distinction in their release notes. Five-inch touchscreen, balanced stereo TRS outputs, serial/parallel loop, 5-pin MIDI and Bluetooth audio. NAM loads to the amp module only. (Its XLR socket is a microphone input, not an output — retail listings frequently get this wrong.)

B Tier — Native, But Reduced I/O

HeadRush Flex Prime. The most interesting entry. It runs the identical NAM engine to the Prime and Core — full native A2-Full, A2-Lite and A1, with the same onboard TONE3000 browsing. So why not S Tier? No XLR outputs, three footswitches, 1/8" MIDI. The tone is S Tier; the back panel is not. If you record at home and play smaller stages, this is arguably the smartest buy on the whole list.

C Tier — Conversion-Based, Decent Hardware

Mooer GE300. Genuinely excellent live hardware — dual XLR outputs, stereo loop, ten footswitches, MIDI in/out/thru — but it converts NAM files into its own format via a built-in converter, and it is A1-era. S-Tier back panel, converted tone.

Valeton GP-200. Converts to Valeton's proprietary SnapTone format, A1 only. Strong I/O for the money.

Mooer GE150 Max. Imports via Mooer Studio, with the architecture never officially stated. XLR outputs, but no FX loop and no MIDI.

Important if you own a GE300 Lite: it has no NAM support at all. It did not receive the GE300's update. It is a different pedal, despite the name.

D Tier — Conversion-Based, Entry-Level Hardware

Mooer GE150 Plus / Pro. NAM import via Mooer Studio, but no XLR, no FX loop, no MIDI, and three to four footswitches. Practice and demo territory.

Sonicake Matribox (QME-50) and Matribox II (QME-100). Sonicake's "Clone" function converts A1 into their own format. On the original Matribox, enabling it disables the cabinet module entirely. No XLR outputs.

To be fair to these units: they are legitimately good pedals for the money. The tier reflects their NAM implementation, not their overall value.

If you want NAM at maximum quality, the place to run it is still a computer — and there is a strong free option.

Two Notes GENOME is a complete amp modeller, effects suite and virtual cabinet engine that supports NAM captures, handled through its CODEX engine. GENOME Intro is free, with a free lifetime licence simply for registering an account. Better still, if you own Two Notes hardware — a Torpedo Captor X, a ReLoad II, an Opus, a ReVolt — you are eligible for a free lifetime GENOME licence. Buy the hardware, the software comes with it.

Practically, that means your NAM captures can run at full A2-Full quality in your DAW, alongside Two Notes' DynIR cabinet system, without spending anything extra on software.

Two clear directions are emerging, and they are worth understanding before you spend money.

First, capture quality is becoming free and universal. The A2 result broke the assumption that the best-sounding models must be proprietary and expensive. When an open-source format outperforms paid systems and runs on cheap chips, the competitive advantage moves away from "who has the best amp model" toward hardware quality, I/O, user interface and workflow. Expect more manufacturers to support NAM natively rather than fight it — and expect pressure on the premium modeller market, including flagship units, to justify their price on features rather than tone alone.

Second, static captures are the current limitation, and the industry knows it. A NAM capture is a photograph of an amp at one setting. Turn the virtual knobs and you are applying EQ, not changing how the circuit behaves. The next frontier is multi-parametric capture — models that reproduce how an amp responds across its entire control range, so the gain knob behaves like a real gain knob.

That work is already happening. Two Notes' AmpNet system, running in GENOME 2.0's PARADEX component, is explicitly built to do this: unlike conventional static captures such as NAM, AmpNet captures aim to reproduce the amplifier's response across the full range of gain and tone controls, bridging classic profiling and flexible modelling.

So the honest forecast: capture accuracy is close to solved and getting cheaper. Interactivity is the next battleground. If you are buying now, buy for the hardware and the workflow, because the tone technology is going to keep improving underneath you — and with NAM, those improvements arrive as free firmware updates rather than a new purchase.

NAM is the most significant thing to happen to amp modelling in years, and the reason is simple: it delivers captures of real gear, made from actual sources, at an accuracy that beat the commercial competition in a blind test, for free. Add a library of hundreds of thousands of community captures and the ability to profile your own rig, and it becomes genuinely hard to argue with.

Just make sure the pedal you buy runs it properly. Native beats converted, A2 beats A1, and none of it matters if the back panel cannot get you to front of house.