
Model Change Report
Model Change Report: Suno v5.5, retested
We re-ran 10 recipes from our library against v5.5 and documented what improved, what regressed, and which prompts need rewriting. With sources.
SunoFetch Lab · 2026-08-08 · 9 min
Suno announced v5.5 with claims of improved arrangement coherence and vocal realism (see the official announcement, linked in our Rights and Model Tracker with date). Marketing claims are not measurements, so we re-ran ten recipes from the Evidence Lab against the new build, same prompts, same batch sizes, and compared keeper rates against our v5 baselines.
- Recipes retested
- 10
- Improved keeper rate
- 5 (arrangement-heavy recipes gained the most)
- Unchanged
- 3
- Regressed
- 2 (both regional-instrumentation recipes)
- Prompts requiring rewrites
- 3
“Every model change is a silent edit to your entire prompt library. Retest before you trust.”
The regressions matter most. Both Noor Baig's ornament-survival recipe and the desi drill recipe lost keeper rate on v5.5: sarangi renders drifted toward synthetic portamento more often than on v5. If your work depends on regional instrument authenticity, keep your v5 keepers archived and budget extra runs.
The wins: long-form arrangement coherence is real. The title-theme recipe's 'climax sags at 50 percent' failure mode dropped from 1-in-4 to roughly 1-in-10 in our runs.
Method note: our batches are 10 to 24 runs per recipe, which is enough to see direction but not enough for precision. Treat our percentages as weather, not physics. Full run logs are attached to each updated recipe page.
Independent educational resource. Not affiliated with or endorsed by Suno, Inc. Suno is a trademark of Suno, Inc.