● ModelsNov 6, 2026 · 05:381 source

Marin's Mochi 535B Open Model Finishes Pretraining on 18T Tokens

From marketbrief.now

AI;DR

The Mochi 535B-A23B open source model has completed its pretraining phase after processing 18 trillion tokens and will now undergo long context training and supervised fine tuning before public release.

Read originalSign in to suggest

Sources

#llmMarkdownScore 4.04.2k tokens
Image: marketbrief.now
Read original