xAI Ships Grok 4.6: Bets on Post-Training Over Scale-Up, Halves Pricing
eyishazyer · x · 2026-08-13
xAI has officially shipped Grok 4.6. According to analysis, the model retains the same 1.5T base model as its predecessor 4.5 without scaling up, routing all performance gains entirely through post-training techniques like RL and fine-tuning.
The model is now live across its API, Cursor, OpenRouter, and other platforms. It is priced at $2/$6 per million tokens (input/output), less than half of what most frontier labs charge. xAI claims it matches GPT-5.6 Sol with a score of 61 on the AA Intelligence Index. However, the widely circulated 1753 ELO score is Musk's internal number, with independent verification from LMArena and Artificial Analysis still pending.
Related event: xAI Launches Grok 4.6 with Top-Tier Performance at Lower Cost(59 posts)→
More from Models
- Ollama Confirms Computer Use Feature is in the Works — reach_vb · 2026-08-13
- Developer Predicts Grok Will Win Claude Users Citing xAI's Insane Shipping Speed — XFreeze · 2026-08-13
- AMD Announces Day 0 Support for Qwen3.8-2.4T Open-Weight Model — AccBalanced · 2026-08-13
- New Model Scores 55 on Rescaled AA Benchmark, Competitive but Disappointing — teortaxesTex · 2026-08-13
- LM Arena's Auto-Eval Tool Predicts New DeepSeek Model at Rank 41 — Unusual_Guidance2095 · 2026-08-13
- Claude Opus 5 Tops InferenceBench with 8.90x Speedup via Adaptive Serving — maksym_andr · 2026-08-13