Elon Musk Says He Does Not Care About Benchmark Scores as Grok 4.8 Remains Unreleased

Grok 4.8 has not been released, and Elon Musk said he does not care about benchmark scores, emphasizing the model’s practical uses instead. According to ChainCatcher, Musk said on October 30 that development had taken longer than expected but promised the new model would not disappoint. On September 14, Musk said Grok 4.8 has 2.5 trillion parameters and uses a new C++ training system. He expected pretraining to finish by the end of that week, followed by reinforcement learning. More than six weeks later, the model is still unreleased. Artificial Analysis found that, at the same xhigh reasoning level, Grok 4.7’s overall intelligence index was two points higher than Grok 4.6’s, rising from 44 to 46. Average output per question increased from about 38,000 tokens to 81,000. Some Cursor community users have also complained that Grok 4.7 uses their quotas faster without delivering a noticeable improvement. In a September interview with CCTV, Musk acknowledged that Grok 4.7 was behind Claude Opus 5.5 and said it might not catch up with leading models until 2027. In October, he proposed having Grok Bot call external models, including Claude, depending on the task, and said Grok 4.8’s high-speed version would handle many simple requests.