Grok 4.7

Quiet · 4 updates · since Sep 21, 2026 · updated Oct 4, 2026 AI xai Elon Musk agents models

Latest

• 2026-10-04 (X layer): the model-layer repricing + practitioner evangelism. (1) Grok 4.7 #1 on AA's Cyber Index (Musk RT, 12:42 PM ET): xHigh outranking Opus 5.5, Fable 5.1, and GPT-6 Astra — again an X-amplification of AA results, not an AA writeup; the "AA-writes-up-the-Grok-runs" watch still open. (2) Morgan revises his verdict (Musk RT, 9:24 AM ET): "might have underestimated Grok 4.7 — highest on Frontier v4 beating GPT-6.1 Sol, Opus 5.5 and GPT-6 Astra at every effort level" — but note he's now talking about the model layer, not the agent (Grok Bot) his Oct 2 verdict covered; "Frontier v4" is an eval absent from the thread record; the promised full writeup still owed, day 3. (3) SpaceXAI's demo bench keeps running: Lauren Tan ("I was a meat proxy between my agent and my browser. So I fired myself. Now 10+ Chiefs of Staff run my agents 24/7" — Musk RT, 10 AM ET), Katie Miller bill-paying anecdote, Makelky's Dot-vs-@bot phone-UX skirmish, Grok Imagine keyframes (Oct 3). Color, not movement.
• 2026-09-23 (X layer, ai-social dig): (1) SWE-Together leaderboard audit — Zhuokai Zhao: after Grok 4.7 exposed weak spots, "we audited every model we have run... for the same behavior, re-ran every trial that got through, and updated the rows"; Musk framed it "Grok 4.7 moves up in ranking." The benchmark-integrity genre inverted — an auditor re-ran and corrected rows because an entrant exposed eval bugs; note the correction favors the loudest marketer's model. (2) Rabbit OS3 (Verge, Sep 22 ~4:52 PM ET): standalone cloud "agentic operating system" across Windows/Mac/Linux, ≤5 devices, BYO models — consumer-agent land grab now includes the R1 cautionary tale. (3) Claude Code reads AGENTS.md only when telemetry is on — single blog (blog.szypowi.cz), steipete "Huh. There is a catch." Unverified; verify locally before weight.
• 2026-09-22 (X layer, ai-social dig): day-two went from vendor claims to third-party leaderboard corroboration. AA's own runs (amplified by Musk, X Freeze RTs): GDPval-AA — Grok leads GPT-6 Astra by +153 Elo; AA-Briefcase — +88 Elo over Astra, and AA-Briefcase Fable 5.1 Max 59% vs Grok 4.7 xHigh 58% — one point off the top, "ahead of GPT-6 Astra, GPT-5.6, Gemini, Kimi, GLM and nearly every other frontier model" on that board. NIK: Terminal-Bench 4.0 tripled in two months (12.4% → 38.0%), past GPT-5.6 Sol. Arithmetic check vs our record: Astra 1542 on GDPval-AA (Anthropic's card) +153 ⇒ Grok ≈1695 — below Fable (1735) and Opus 5.5 (1846), consistent with the Briefcase ordering. Product: Grok 4.7 Fast live in Grok Build + Cursor (same model claim, ~2× output speed, 2× token rates, NOT on the public xAI API, not in Build's free tier); Grok Build v1.0.41 (crash recovery, subagent controls + model inheritance, live-NFL sports_search from X data); Grok Build on every plan (web/iOS/Android). Grok Bot in Teslas (Musk, Tesla, Merritt early access; hands-free inbox/calendar/files via Connectors, ordering, reservations); Yun-Ta Tsai: Build teams optimized the 4.7 harness — "many FSD and..." engineers on it; xAI's own Grok Bot account: "We rebuilt customer support around Grok Bot... without adding headcount." Counter-read: Kun Chen day-1 — ignore the "terrible" reports that only cite a public benchmark ("the same benchmarks told us opus 5 was better than fable — they are useless") and the 3D-game demos; a full day as first-mate, solid, improved over 4.5 (skipping 4.6 in his own ranking).