Elon Musk Says xAI's Grok 4.8, a 2.5-Trillion-Parameter Model on New C++ Stack, Will Finish Training This Week, Leapfrogging Delayed Grok 4.7 Release

Elon Musk announced on September 13, 2026, that xAI's Grok 4.8, a 2.5-trillion-parameter model trained on a new C++ software stack, will finish its initial training run this week and enter reinforcement learning, according to posts on X. Musk stated on September 14 that Grok 4.8 will be a noticeable improvement over the still-unreleased Grok 4.7, which he positioned as roughly on par with Anthropic's Claude Opus 5.0. He projected Grok 4.9 as probably Astra-Fable class and Grok 5 as maybe better than anything, framing the roadmap as a sequential climb toward AGI-level capability. Grok 4.8 represents a 67 percent parameter increase over the 1.5-trillion-parameter Grok 4.6 released on August 12, and leapfrogs the delayed Grok 4.7, which missed multiple target dates including one set for September 11. As of September 15, xAI has released no model page, API identifier, pricing, context window, or benchmark card for Grok 4.8, and every specification remains a founder claim rather than a shipped product. The company is training the model on its Colossus 2 supercluster, described as the world's first gigawatt-scale training facility, and reportedly has multiple foundation models training concurrently with variations potentially reaching 10 trillion parameters. For the AI development community, Musk's announcement of a model two versions ahead of the current release illustrates both xAI's aggressive compute deployment and the unpredictability of its public release cadence, as Grok 4.7 remains unreleased despite multiple prior commitments. The shift to a new C++ software stack suggests infrastructure re-platforming aimed at training efficiency or hardware compatibility, though xAI has not disclosed technical details. The 2.5-trillion-parameter scale and rapid iteration speed position xAI as one of the few organizations pushing parameter counts into multi-trillion territory, a race that requires enormous capital, power, and chip access. Whether the models ship on schedule or