ChatGPT — Cadence Benchmark Assessment

Date: 2026-03-29
Source: ChatGPT (OpenAI)
Context: Realistic performance expectations for BlackRoad's text-to-music tool

---

No Published Benchmarks Yet

No generation times, FAD scores, listening tests, or side-by-side comparisons with Suno/Udio.

Inferred Performance (based on hardware)

Generation Speed

Short clips (5-30s): 5-30 seconds on 1-2 nodes

Full tracks (1-3 min): 30-120+ seconds on basic setup

Iteration: unlimited, no credits — big win over cloud

Quality

Strong: persistent memory for brand-consistent output

Good for: music beds, intros/outros, backgrounds, ideation

Limitations: complex orchestral/vocal may show artifacts until models mature

No public audio samples available

Scalability

Start 1 Pi + Hailo → add nodes for speed

Low power, 24/7 feasible

Zero recurring costs after setup

vs Cloud Tools

| Dimension | Suno/Udio | Cadence |
|---|---|---|
| Speed | Fast (cloud GPU) | 5-30s short clips (edge) |
| Cost | Credits add up fast | Zero after hardware |
| Memory | None (re-prompt every time) | Remembers brand/style |
| Privacy | Uploads required | Fully local |
| Quality | Higher fidelity (big GPU) | Trade-off for ownership |
| Consistency | Generic each session | Brand DNA preserved |

vs Traditional (Logic, Ableton)

Faster for AI-assisted generation/beds

Likely needs DAW export for pro mixing/mastering

Honest Bottom Line

> "Cadence looks promising as a sovereign, memory-driven music ideation and bed generator. On a 5-node Pi fleet, it should deliver responsive performance for typical creator needs with excellent cost/privacy wins."

> "However, without published benchmarks, audio samples, or detailed quality metrics, it's hard to quantify exactly how it performs today versus hype."

---

Raw ChatGPT output preserved verbatim. Filed 2026-03-29.