Quasar's 120B Decentralized AI Model Exposed: Weights Match Existing Model, Token Crashes 70%

0xSammy · x · 2026-08-02

The crypto AI space just faced a major trust crisis. Bittensor's Quasar subnet recently hyped the decentralized training of a 120B-parameter model with a 5M token context window.

However, analysis revealed that 98% of the released weights matched an existing open-source model, raising serious doubts about whether the model was actually trained from scratch or simply copied. Following these allegations, the project's token price plummeted by roughly 70%.

This incident highlights a critical pain point in decentralized AI: verifiable training provenance is rapidly becoming a fundamental requirement. Just as regulatory compliance demands immutable agent logs for institutions, decentralized ledgers might be the key to executing AI training verification.

Related event: Quasar 120B Model Caught Plagiarizing Weights, Token Plummets(3 posts)→

Original post →

More from Fun

Fun channel →