이미지: X — 뉴스 앰프 화면 갈무리
Summary
- Elon Musk posted on X that the upcoming Grok 4.7 model will surpass every model currently available
- Musk claimed SpaceX's training data is so unique that no model will beat 4.7 in real-world engineering
- An AI news account operator, citing the recently released Grok 4.6, said the claim deserves to be taken seriously
- 발언자
- 일론 머스크(Elon Musk), xAI
- 게시 시각
- 2026년 8월 12일 오후 8시24분(현지시간), 조회수 10만4700회
- 핵심 주장
- Grok 4.7이 현존 모든 모델을 능가할 것
- 근거로 제시한 자산
- 스페이스X(SpaceX)의 실제 엔지니어링 훈련 데이터
- 직전 모델 성과
- Grok 4.6, Artificial Analysis Intelligence Index 61점으로 GPT-5.6 Sol과 동급
The next card Musk has teased
Elon Musk posted a short message on X. It stated that the next model, Grok 4.7, would "surpass every model out there right now." He acknowledged Anthropic as a good company that would soon release an improved model, but immediately added a caveat: "SpaceX's training data is so unique that I'd honestly be surprised if any model beats 4.7 in actual engineering."
The operator of an AI news account who quoted the post said they are "now taking this seriously," calling the recently released Grok 4.6 "the leap we were hoping for." They added that if a model estimated to have around 10 trillion parameters is still yet to come, Musk's remarks might not just be bluster. However, the same post also mentioned that Anthropic is reportedly preparing a follow-up model under the codename "Fable 5.5," which appears to be speculation rather than an official announcement.
What this is about
Grok 4.6 was released on August 12 and drew significant attention. On the Intelligence Index compiled by evaluation firm Artificial Analysis, it scored 61 points, placing it in the same top tier as GPT-5.6 Sol and marking a 5-point improvement over its predecessor, Grok 4.5. On practical benchmarks like GDPval-AA v2, it ranked second behind Claude Opus 5, and it is reportedly already being tested within the coding tool Cursor. Pricing remains the same as the previous model, but the cost per task is said to be much lower.
The "SpaceX training data" Musk emphasized appears to refer to real engineering records such as rocket design, launches, and orbital calculations. Unlike ordinary text or code data, this is data bound by physical laws and real-world constraints, which xAI seems to be presenting as a point of differentiation. However, no verified figures have been provided showing how much such data actually contributes to model performance.
So what changes
What has emerged so far is not an official benchmark or a launch announcement, but a confident tease from a CEO. Given that Grok 4.6 achieved top-tier results just over a month after release, it does seem clear that xAI is preparing its next model on a short cycle. With competitors like Anthropic and OpenAI also releasing follow-up models at a similar pace, if Grok 4.7 actually materializes, the competition at the top could be reshuffled once again.
