Bittensor Subnet Claims Decentralized Training of 120B Model with 5M Context, Yet to Be Verified

markjeffrey · x · 2026-08-01

Bittensor subnet Quasar subnet 24 claims to have decentrally trained a 120B parameter model with a 5 million token context window. The model, named Quasar-120B-Preview, is slated for release on HuggingFace, but as of now it has not been uploaded, so the claims remain unverified. The release will include full weights, architecture configs, tokenizer, and other files.

Related event: Bittensor's Quasar Claims Decentralized Training of 120B Model(2 posts)→

Original post →

More from Models

Models channel →