Elon Musk recently acknowledged that the upcoming Grok 4.8 model from xAI will not outperform Anthropic’s Claude Opus 5.5. Musk described the rival model as a masterpiece, while noting that Grok 4.8 is not yet at that level. Anthropic released Opus 5.5 on September 22, reporting a 65.76% score on Terminal-Bench 4.0, which exceeds the 57.9% score of OpenAI’s GPT-6 Astra.
Despite the current gap, Musk remains optimistic about the future of his company's technology. He stated that the subsequent version, Grok 5, should surpass Opus 5.5 within two or three months. However, he provided no firm release date for this update, and historical performance suggests these timelines may be subject to change.
The development of Grok 4.8 involved significant technical shifts. The xAI team completed pretraining and is currently in the reinforcement learning phase. Developers replaced the JAX and SGLang systems with a new architecture written in C and C++. Musk credited Starlink engineers for their contributions, noting that the stack incorporates practices from SpaceX, such as custom headers and precise time calculations.
The company is currently finalizing the model after spending time cleaning up supervised fine-tuning data. Musk described the foundation of Grok 4.8 as solid and announced that a soft rollout to early users is expected later this week. Anthropic has not issued a public response to these statements regarding the competitive landscape of the AI industry.
