Michele Catasta

@pirroh

President & Head of AI @Replit | 🇮🇹 @🇺🇸

following
followers

Replit Agent's reported gains over a single-worker setup on coding benchmarks

Replit says its Agent lets the core model choose and reuse subagents as work unfolds. In its tests, it reports 11- and 16-point gains over a single-worker setup on DeepSWE and Terminal-Bench, respectively.

4 Sources, 21d ago, updated 21d ago

Featured in

Replit Agent's reported gains over a single-worker setup on coding benchmarks

Featured in

Replit Agent's reported gains over a single-worker setup on coding benchmarks