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.
More from Models
- Gemini 4 tops public coding benchmarks but drops to bottom on private tasks — Hesamation · 2026-10-30
- Your AI agent would cost 260% more without prompt caching, Arena data shows — arena · 2026-10-30
- EachLabs launches Lockstep recast model claiming 75% cost cut and better quality — miilesus · 2026-10-30
- ChatGPT cited a 300+ page government legal PDF it retrieved on its own — lilyraynyc · 2026-10-30
- Best MLX model for French-to-English translation on an M2 Mac? — greenreddits · 2026-10-30
- AI-driven cyber ops near the sophistication peak set by GTG-1002's Claude Code campaign — Jsevillamol · 2026-10-30