Limite 1B maker says outputs average ~50k tokens incl. CoT, advises longer generation on AIME

tensorqt · x · 2026-09-23

After GiuseppeSaccar3 benchmarked Paradigma's Limite 1B via an experimental MLX port on an M4 Pro, tensorqt (apparently from the team) replied that in their evals Violetto averages 50k output tokens including CoT, suggesting longer generation — explaining why many local AIME runs hit the 8192-token cap.

Related event: Limite 1B runs locally on M4 Pro at ~57 token/s, scoring 4/8 on AIME sampling(2 posts)→

Original post →

More from Models

Models channel →