AI music generators do not stand still. New versions can arrive before the industry has even agreed on how to describe their sound, which makes a static detector less useful with every release.
The update
ArtifactNet’s current validation range now includes Suno v6 and ACE-Step 1.5. The updated detection model is live now for every ArtifactNet account.
That means there is nothing to install, migrate or reconfigure. The same upload flow now uses the newly validated model generation automatically.
What changed
We expanded the validation coverage behind ArtifactNet to include recent output from Suno v6 and ACE-Step 1.5. New coverage is evaluated before it becomes part of the model update, so a generator name is not simply added to a product page; it becomes part of the evidence used to judge whether the detector is ready to ship.
The release also updates the main detection model and its matched review layer together. Keeping those pieces aligned matters because an AI-music result should be judged by the full system, not by one isolated score.
What the latest benchmark showed
We re-ran our current ArtifactBench evaluation against the preceding production model. The new model improved both detection quality and the rate at which ordinary music is incorrectly identified as AI-generated.
| Metric | Previous model | Current model |
|---|---|---|
| F1 | 0.9933 | 0.9963 |
| Aggregate false-positive rate | 1.22% | 0.37% |
Those numbers are one checkpoint, not a finish line. The goal is to keep validation current as generator families and distribution patterns change, then make the improved model available without asking customers to change how they work.
Why current coverage matters
For a label, distributor or creator, a detection result is useful only if it reflects the audio that is actually arriving today. Adding a new generation to the validation range helps keep reviews, catalog checks and internal research from being anchored to older examples of AI music.
ArtifactNet is designed for that continuing work: assess the tracks in front of you now, update the evidence as the field moves, and keep the product interface stable while the model improves underneath it.
Try the updated model
Upload a track to run it through the current ArtifactNet detection stack.
Analyze a track →Explore ArtifactNet