Grok 4.7 Misses Target Again; 2.5T-Parameter Grok 4.8 Finishes Training

eyishazyer · x · 2026-09-15

Grok 4.7 missed its September 12 target again — Musk says RL may have over-penalized long responses, making the model give up early. Meanwhile Grok 4.8, a 2.5T-parameter model on a new C++ stack, finishes training this week and enters RL, with 4.9 and Grok 5 queued behind it.

Original post →

More from Models

Models channel →