Elon Musk had a rare moment of praise for a competitor on Tuesday, and it came with a blunt caveat about his own product. Asked on X how Grok 4.8 would compare with Anthropic’s Claude Opus 5.5, the xAI founder said it would be ‘not as good.’ In the same reply to user Dimdv, he called Opus 5.5 ‘a masterpiece.’
Musk did not leave it there. He said Grok 5, the next model after 4.8, ‘should exceed Opus 5.5,’ perhaps in two or three months. That timeline is worth treating with caution. He has missed release targets before, and he gave no firm date.
Why Grok 4.8 is late
The exchange started with a technical update. Musk said Grok 4.8 had finished pretraining but was still in reinforcement learning, where a model’s answers are refined with feedback and rewards. He said the team also spent significant time cleaning supervised fine-tuning data, the examples used to teach instruction-following.
Musk said xAI dropped JAX and SGLang, software used for machine learning and serving models, for an inference and reinforcement-learning system rewritten in C and C++. Inference is the step that generates an answer after a prompt. Rewriting that layer can give engineers more control, though Musk shared no numbers showing the gain.
Developer tetsuo responded with ‘C wins again.’ Musk clarified that the code was ‘C with a little C++.’ He also credited Starlink engineers with helping build the Grok training and inference stack. He pointed to SpaceX’s habit of writing its own headers and handling precise time calculations, including leap seconds.
Musk described Grok 4.8 as ‘not yet incredible’ but said its foundations were finally built correctly. He expects a soft rollout to Bot users later this week, an early test before any wider release.
Anthropic’s lead is real
Anthropic released Opus 5.5 on September 22. It performs comparably to the higher-end Fable 5.1 on most work while costing 40% less to run than the previous Opus version on typical workloads, according to the company. The model is aimed at difficult programming jobs and multi-step agent tasks with limited human help.
On Terminal-Bench 4.0, which tests whether agents can finish complex command-line tasks, Anthropic reported 66.4% for Opus 5.5 and 57.9% for OpenAI’s GPT-6 Astra. Grok is far behind on the leaderboards. Musk has not published Grok 4.8 scores that would show the size of the gap.
For now, his target is the limited rollout this week. The promised Grok 5 challenge to Opus 5.5 is still two or three months away by his own estimate.
The post Musk Calls Claude Opus 5.5 a Masterpiece, Says Grok 4.8 Falls Short appeared first on TheCryptoUpdates .
