No, Laya isn't capped at 512 tokens — it's ModernBERT with 8192-token configs

antoine_chaffin · x · 2026-09-21

Responding to rumors that Laya only supports 512-token context, JF Puget shows the encoder is ModernBERT with max length 8192 in both tokenizer and model configs on Hugging Face. Antoine Chaffin adds that ModernBERT (and thus Ettin & mmBERT) has long shown strong long-context performance while scaling efficiently to long sequences.

Original post →

More from Models

Models channel →