NAM vs ToneX vs Neural Capture: What a Thousand Blind Listeners Actually Picked

September 2, 2026by OpenFret Team

Amp capture used to be a settled argument decided by brand loyalty. Then someone ran an actual test, and the free format won it, which is not how these stories usually go.

The four things being compared

They all answer the same question, which is how to make a computer behave like one specific amplifier instead of a generic idea of one.

Neural Amp Modeler is the open-source one. Steve Atkinson released it in 2022, anyone can capture an amp with free tools, and the resulting .nam file runs in dozens of hosts on every platform. There is a fuller explanation of how it works if you want the mechanics.

ToneX is IK Multimedia's. It has a free tier, paid tiers above it, a pedal, and ToneNET, a library of user models. It is the most consumer-friendly of the group and the one most likely to already be on a beginner's laptop.

Neural Capture belongs to Neural DSP and runs on the Quad Cortex and in their plugins. It is the only one on this list that captures an amp without a computer in the room, which is a genuinely nice thing to have at a rehearsal.

Proxy is Line 6's capture technology, the newest name here, and the one most tied to its own ecosystem.

The test

TONE3000 ran it for the launch of NAM A2 in June 2026, in two halves.

The measured half took 39 tones and scored every engine against the real recorded gear on ESR, MAE, LOG_MEL and MRSTFT, then expressed the outcome as Bayesian Elo ratings so the numbers could be compared across tones rather than cherry-picked one at a time.

The listening half used MUSHRA, the standard method for this kind of question, where you rate candidates against a hidden reference on a scale of 0 to 100 without knowing which is which. More than a thousand people took part and submitted over a hundred thousand ratings. NAM A2 took the highest median ratings, ahead of Neural Capture V2, ToneX V2 and Proxy, and the gap was not subtle.

The conflict of interest, stated plainly

TONE3000 built A2 and TONE3000 ran the test. That is a reason to be careful, not a reason to dismiss it. They published the test code and the raw ratings, so anyone who wants to re-run the analysis can, and in a year of people trying, the result has held up. Compare that with the usual approach to modelling claims, which is a sentence on a product page and a demo video recorded by the company.

What the test does not tell you

It measures the engine, not the capture. That distinction is everything in practice.

A capture made in a treated room with a good mic on a well-maintained amp, at a level that suits the format, will beat a careless capture of nicer gear in any engine. Most of the variation you hear when you browse a library is people, not maths. This is the main reason a reputable capture author is worth more to you than the format war.

It also does not measure the parts you live with: how quickly you can find a sound you like, whether the thing runs on a stage without a laptop, whether the interface annoys you at eleven at night.

So which one should you use?

If you practice and record at a computer, NAM, and specifically the TONE3000 plugin, which is free, runs standalone, and browses more than 700,000 tones from inside the plugin. There is no argument on value and, on the published numbers, none on accuracy either.

If you play out, hardware wins on reliability alone, and the Quad Cortex remains the unit to beat. Its CorOS 4.1.0 update from August 2026 was the biggest one it has had. Note that NAM is arriving in hardware too, since A2-Lite runs on cheap chips and Blackstar, Darkglass, HeadRush and others have shipped support.

If you already own ToneX and like it, keep using it. The measured gap is real and it is also smaller than the difference between two mic placements on the same cab. Nobody at your gig is going to identify your capture engine.

The part that decides how you sound

Every engine here reproduces what you play with more accuracy than most players can use. A capture that resolves the bloom of a tube amp will resolve a rushed sixteenth just as clearly, and the audience only hears the second one.

Whichever you land on, put something in the chain that checks your playing rather than your tone. Guitar Quest listens through the same interface the amp sim is running on and grades note by note, so the hour you were going to spend auditioning captures turns into an hour of practice you can see the results of. It is $30 once, and the demo runs in the browser without an account. Same idea whichever engine you settled on: