xAI 发布 Grok 4

2025 年 7 月 8 日，xAI 发布了 Grok 4。对大多数人来说，这个消息的冲击不在于又有一个新模型，而在于一个事实：马斯克手下的模型，第一次真正站到了 GPT、Claude、Gemini 同一张桌上，成了推理旗舰的正式玩家。xAI 不再是「马斯克的另类项目」了。

回到 2023 年，xAI 成立时，Grok 的定位充满了个人色彩——「追求真相、少设限」，风格像它的老板。但早期 Grok 的能力和头部有真实差距，更多人把它当成噱头。xAI 这几年做的事其实很朴素：疯狂堆算力、快速迭代，然后让时间验证这条路。

Grok 4 正是这条路的成果。它在推理、代码、多模态上大幅跃升，多项评测进入了第一梯队。配合 X（推特）的生态整合，它在实时信息与开放语料上有独特位置。发布前后，xAI 完成巨额融资、估值飙升——资本用真金白银投了票：马斯克这套「算力换时间」的打法，确实追上了。

Grok 4 的意义超越了 xAI 本身。它改变了头部阵营的格局：之前是三家你追我赶，现在至少是四家。多一个重量级玩家，意味着整个行业的迭代更快、定价更合理、开发者选择更多。它也用实例回答了那个悬而未决的问题——后来者能不能靠砸算力追平先行者？至少这一次，答案偏向「能」。

回看 Grok 4，它最值得记录的是一件事：竞争的天花板被抬高了。当一家敢砸千亿美元算力的公司成为常驻玩家，头部模型之战就不再是温和的寡头游戏。Grok 4 让这场游戏更激烈，也更有意思。

On July 8, 2025 xAI released Grok 4. For most people the impact wasn't another new model—it was the fact that Musk's model had, for the first time, taken a seat at the same table as GPT, Claude, and Gemini, becoming an official player in reasoning flagships. xAI was no longer "Musk's side project".

Back in 2023, when xAI was founded, Grok's positioning was full of personality—"pursue truth, few limits", styled after its owner. But early Grok genuinely lagged the leaders in capability, and many treated it as a gimmick. What xAI did over the following years was plain: throw enormous compute at the problem, iterate fast, and let time validate the path.

Grok 4 was the fruit of that path. Reasoning, coding, and multimodality jumped, and multiple evals put it in the top tier. Combined with integration into X, it held a distinctive position on real-time information and open data. Around the launch xAI closed a massive funding round and its valuation soared—capital voted in hard currency: Musk's "buy time with compute" play had caught up.

Grok 4's significance went beyond xAI itself. It reshaped the top-tier landscape: what had been three companies trading blows became at least four. One more heavyweight means faster iteration across the industry, more rational pricing, and more choice for developers. It also answered a lingering question—can a latecomer catch the leaders by pouring in compute? This time, the answer leaned toward "yes".

Looking back at Grok 4, the most worth recording is a single fact: the ceiling of competition was raised. When a company willing to spend tens of billions on compute becomes a permanent player, the flagship war stops being a gentle oligopoly. Grok 4 made the game more intense—and more interesting.