Musk reveals Grok 4.8: entire RL and inference stack rewritten from scratch in C/C++
elonmusk · x · 2026-11-10
Elon Musk shared a major update on Grok 4.8: the engineering team dumped JAX and SGLang and rewrote the entire RL and inference stack from scratch in C/C++, which took extra time.
Key details:
- Pre-training is complete, with heavy cleanup needed on SFT data
- The model is now in reinforcement learning, and eval scores are starting to climb
- A soft rollout to Bot users is planned for later this week
Musk added that Grok 4.8 will "bear no resemblance to 4.7" and that while "it is not yet incredible," the foundations are built right this time.
More from Infra
- 10,000-hour vision-tactile dataset drops as robot data gold rush fuels shovel sellers — CyberRobooo · 2026-11-10
- Stratechery: Jensen Huang's Story Worsens — Governance Risk Looms Larger Than National Security — Stratechery · 2026-11-10
- Running open-weight models locally: Kimi K3.1 needs 594GB, ~$38k for 15 tok/s — ravikantagrawal · 2026-11-10
- InfraRSI Applies Recursive Self-Improvement to Training Infra: 2.75x Speedup, MFU Lifted from 17.4% to 30% — TokenRhythm · 2026-11-10
- ClickBench Playground Lets You Run SQL Against 110+ Databases Side by Side — mgill25 · 2026-11-10
- 4x RTX 3090 vs one RTX 5090: Malaysian builder weighs VRAM against newer architecture — whity2773 · 2026-11-10