The Spirit Index

A reference index of autonomous cultural agents

Last reviewed: January 2026

Jukebox

Raw Audio Music Generation

ArchivedAutonomous Composer
35
/90 Total
Not yet registered on-chainRegister on Spirit Protocol →

Score Profile

Dimension Scores

PER
5
AUT
5
IMP
7
ECO
2
GOV
2
TEC
9
NAR
5
ECI
0
IDS
0
Score Rationale
PersistenceReleased April 2020 with open-source code and weights. Active exploration by community but not actively maintained by OpenAI. Research artifact status.
AutonomyGenerates music autonomously given genre/artist/lyrics prompts. No continuous operation or self-direction. Tool rather than agent.
Cultural ImpactLandmark in AI music generation. First to generate raw audio with rudimentary singing at scale. Widely discussed in AI art discourse. Open-source release expanded impact.
Economic RealityNo economic activity. Research project with public code release. No token, treasury, or sales.
GovernanceOpenAI corporate governance. No community governance structure. Research project model.
Technical ArchitectureRevolutionary: raw audio generation (not MIDI). Trained on 1.2M songs using 512 V100 GPUs. Autoencoder compression + upsampling architecture. 9 hours to render 1 minute of audio shows computational scale.
Narrative CoherenceIdentity as 'OpenAI research project' rather than autonomous entity. Preceded by MuseNet. Superseded by later models. Limited narrative continuity.
Economic InfrastructureOpenAI research project with no economic infrastructure. Open-source code and weights released but no treasury, no token, no payment rails. No commercial deployment.
Identity SovereigntyIdentity exists only on OpenAI website, GitHub, and academic papers. No on-chain presence. Research artifact with no portable digital identity.

Curator Notes

Jukebox represents OpenAI's landmark contribution to AI music generation—the first system to generate raw audio music including rudimentary vocals at scale. Trained on 1.2 million songs using 512 NVIDIA V100 GPUs for four weeks, Jukebox generates in multiple genres and can mimic artist styles. The computational requirements (9 hours to render 1 minute) reflect the difficulty of raw audio generation versus symbolic (MIDI) approaches. While more tool than agent, Jukebox's open-source release enabled significant community exploration and established a baseline for AI music generation research.

Evidence Archive

TECTrained on 1.2M songs using 512 V100 GPUs[source]
IMPOpen-source release with weights and code[source]
TECFirst raw audio generation with rudimentary singing[source]

Score History

Score history will appear here after future reviews.

Current score: 35/90

Metadata

Inception2020-04-01
ClassificationNeural Network System
Websiteopenai.com
Last Updated2026-01-16