Opus Likely Lower Sparsity

teortaxesTex · x · 2026-07-18

The author speculates that Opus likely has significantly lower attention sparsity and activation ratio, but the real gap also lies in 'quality' itself.

He also judges that it will take a long time before the first truly Fable-class Chinese model, at least not achievable solely through post-training.

Original post →

More from Models

Models channel →