SIAgentNews.com
← All news

Musk admits Grok 4.8 will trail Anthropic's Claude Opus 5.5

Elon Musk acknowledged that xAI's upcoming model falls short of its rival's performance but predicts a reversal within months.

· First reported by 디지털투데이

40 percentOperating cost reduction for Claude Opus 5.5

The short version

  • Musk confirmed Grok 4.8 has finished pre-training and is currently in the reinforcement learning phase.
  • xAI is replacing JAX and SGLang with a custom system written in C and C++ for inference and training.
  • Anthropic's Claude Opus 5.5 scored 66.4 percent on the Terminal-Bench 4.0 evaluation for complex tasks.
  • Grok 4.8 is scheduled for a limited soft rollout to some bot users as early as this week.

Why it matters

This admission provides a rare moment of transparency regarding the current hierarchy of AI capabilities. It signals that while xAI is struggling to beat the current leader, they are aggressively rebuilding their underlying software architecture to close the gap quickly.

Admitting the current leader

Elon Musk recently offered a rare moment of humility regarding the state of artificial intelligence. He acknowledged that xAI's next major model, Grok 4.8, will not outperform Anthropic's Claude Opus 5.5. Musk even went as far as to describe the rival model as a "masterpiece."

The admission comes as the AI industry races to develop models capable of handling complex, multi-step tasks. While xAI aims to compete with the biggest names in the field, Musk admitted that Grok 4.8 is "not" at an amazing level yet. This transparency highlights the significant gap that still exists between the leading models and the newer entries in the market.

Anthropic released Claude Opus 5.5 on September 22, positioning it as a high-performance tool for programming and agent tasks. The model is designed to deliver performance close to the top-tier Fable 5.1 while significantly reducing costs. Musk's comments suggest that Anthropic currently holds a distinct advantage in both capability and efficiency.

A total software overhaul

To bridge the performance gap, xAI is undergoing a massive overhaul of its technology stack. The company is discontinuing the machine-learning frameworks JAX and SGLang, which were previously used to run and serve AI models. In their place, they are rewriting their inference and reinforcement-learning systems from scratch using C and C++.

Inference is the specific process by which an AI system generates a response to a user's prompt. By moving to a lower-level programming language like C++, xAI aims to optimize how these models function. Musk described the new code as "a bit of C++ added to C," indicating a move toward a more streamlined and efficient foundation for their software.

This engineering effort is not happening in isolation. Musk noted that engineers from Starlink contributed to building the training and inference systems for Grok. This collaboration suggests that xAI is leveraging its broader corporate resources to ensure the underlying structure of the next model is built correctly.

Training and rollout plans

Grok 4.8 has officially completed its pre-training phase and is now moving into reinforcement learning. This is the stage where the model learns to follow specific instructions and human preferences. To ensure the model behaves as intended, xAI has also overhauled the supervised fine-tuning data used during this process.

Because the underlying structure has been rebuilt, xAI plans to conduct a limited soft rollout. This means the model will be released to a small group of bot users as early as later this week. This controlled release allows the company to monitor performance and identify issues before the model is made available to the general public.

The company is using this period to verify that the new software architecture produces the desired results. While the official comparison scores for Grok 4.8 have not been released, the company is focused on these internal checks. This strategy allows them to refine the model's capabilities in a live environment before a full-scale launch.

The road to Grok 5

Despite the current limitations of Grok 4.8, Musk is confident about the immediate future of xAI's development. He stated that the follow-up model, Grok 5, is expected to surpass Claude Opus 5.5. He provided a specific timeline for this goal, estimating that the leap could happen in 2 to 3 months.

This aggressive timeline suggests that xAI expects to see continued acceleration in its development cycle. By moving away from existing frameworks and building custom systems, they hope to achieve faster iterations. The goal is to move from a trailing position to a leading one in a very short window of time.

For now, the industry will be watching the soft rollout of Grok 4.8 closely. The success of this limited launch will serve as a litmus test for xAI's new engineering approach. If the new C++ based systems work as intended, it could pave the way for Grok 5 to meet Musk's ambitious goals by the end of the year.

What to watch

  • How Grok 4.8 performs during its limited soft rollout this week.
  • Whether xAI can successfully transition to its new C and C++ infrastructure.
  • The release of official comparison scores for Grok 4.8.

Try it

Check the Terminal-Bench 4.0 leaderboard to see how Claude Opus 5.5 compares to other models like GPT-6 Astra.

Sources and links

First reported by 디지털투데이. Written by the SI Agent News Desk from the reporting above.Written with AI from the source above. Every number is checked against the source. Spot a mistake? Corrections.

Liked it? Say why

Add a line for other readers.

No account needed. We never show your email. Comments with links, ads or abuse are held and not shown. Your comment shows after a quick check.

More AI news