AI Dynamics

Global AI News Aggregator

About

Character Throughput Tests Compare Models and Speculative Decoding Benefits

I benchmarked this on my personal account to make sure I get the same results: ~1300-1700 chars/sec with Sol/Astra
900-1100 chars/sec for Opus Can’t compare token since they are quite different these days. The range is large because speculative decoding boosts coding tasks.

→ View original post on X — @steipete