Grok 4.7 still isn't out. Here's a simple math read on when it might land.
All the dates below are from Elon Musk on X, not from xAI's own docs. As of Sep 17, docs.x.ai still lists Grok 4.6 as the latest flagship.
The public targets moved like this:
• Jul 24: "in 4 weeks" → about Aug 21
• Aug 12: "3 to 4 weeks" → about Sep 2–9
• Sep 2: "in 10 days" → about Sep 12
• Sep 11: "a few more days" → about Sep 14–16
Against the first hard target (Aug 21), we are already ~27 days late — almost one full original promise window. Against the last hard target (Sep 12), elapsed time is 1.5× the promised 10 days.
Prior Grok slips give useful multipliers:
• Grok 4.5: first promise window ~40 days, actual ~89 → ~2.2×
• Grok 4.6: promised 14 days, actual 19 → ~1.36×
Apply those to 4.7:
• Optimistic (4.6-like): around Sep 16–17 — basically now
• Mid (1.5× from the Sep 2 clock): around Sep 17
• Pessimistic (4.5-like from Jul 24): around Sep 24
The successive targets also compressed: Aug 21 → early September → Sep 12 → "a few days." Jump sizes roughly halved each time. That pattern, plus Musk's Sep 11 note that RL may have over-penalized response length (model quits hard tasks early, weak self-check), points to a short remaining window — not a multi-month redo — unless the fix is deeper than he framed it.
My read:
• P50: Sep 20–22
• P75: Sep 24–26
• P90: early October if RL needs a real retrain
On Sep 14 he was still talking about 4.7 as the near-term model (roughly Opus 5.0 class), with 4.8 next. So this does not look cancelled — just late. Until a model page shows up on docs.x.ai, treat every date as aspirational.
Top comments (0)