Limite 1B runs locally on M4 Pro at ~57 token/s, scoring 4/8 on AIME sampling

A mathematician ran Paradigma's Limite 1B (Violetto) locally on M4 Pro via an experimental MLX port at about 57 tokens/s, scoring 4/8 on sampled AIME 2025 problems. An account believed to be official noted the model averages 50k output tokens on AIME and recommends allowing longer generation time.

2026-09-23 ~ 2026-09-23 · 2 related posts

• Limite 1B on M4 Pro via experimental MLX port: ~57 tok/s, 4/8 on sampled AIME 2025 — GiuseppeSaccar3 · 2026-09-23
• Limite 1B maker says outputs average ~50k tokens incl. CoT, advises longer generation on AIME — tensorqt · 2026-09-23