[Digital Today intern reporter Seung-a Yoo] Elon Musk (일론 머스크), founder of xAI, acknowledged that his company’s next artificial intelligence model, Grok 4.8, will not outperform Anthropic’s Claude Opus 5.5. He also called the rival model a "masterpiece."

On Nov. 10 (local time), IT and blockchain outlet Decrypt reported that Musk said on X, formerly Twitter, that Grok 4.8’s performance would be "not as good" as Claude Opus 5.5.

Musk said Grok 4.8 has completed pre-training and is now undergoing reinforcement learning. He added that the supervised fine-tuning data used to train AI systems to follow instructions has also been overhauled. xAI plans a limited soft rollout to some bot users as early as later this week.

Musk said Grok 4.8 is "not" at an amazing level yet, but stressed that the underlying structure was "built correctly this time." That means it will be released first to some users to check performance before the official launch.

He also disclosed an overhaul of the technology stack. xAI is discontinuing the machine-learning framework JAX and SGLang, used to run and serve AI models, and has rewritten its inference and reinforcement-learning systems from scratch in C and C++. Inference is the process by which AI generates an answer to a user’s question. Musk said the code is "a bit of C++ added to C," and added that Starlink engineers also contributed to building Grok’s training and inference systems.

Musk said the follow-up model, Grok 5, should surpass Claude Opus 5.5. He put the timing at 2 to 3 months later. He did not present specific comparison scores or a confirmed release schedule.

Anthropic unveiled Claude Opus 5.5 on Sept. 22. Anthropic said the model delivers performance close to the top-tier Fable 5.1 on most tasks while cutting operating costs by 40 percent compared with the previous Opus model for typical tasks. It focuses on complex programming and multi-step AI agent tasks.

In Anthropic’s Terminal-Bench 4.0 evaluation, Claude Opus 5.5 scored 66.4 percent. OpenAI’s GPT-6 Astra scored 57.9 percent. The evaluation measures how successfully AI agents complete complex tasks in a command-line environment. Meanwhile, the model that posted a low score on the existing leaderboard was not Grok 4.8 but the earlier version, Grok 4.7. Grok 4.8’s official comparison score has not yet been released.

In the near term, the limited launch of Grok 4.8 slated for this week will be the key point to watch. Attention is on how much xAI has improved performance through its new software structure, and whether the Grok 5 that Musk has flagged can surpass Claude Opus 5.5 within 2 to 3 months.

@Dimdv99 @tetsuoai Not as good. Opus 5.5 is a masterpiece. But Grok 5 should exceed Opus 5.5 and maybe that happens in 2 or 3 months, with continued acceleration thereafter.