Grok 4.8 has yet to be released, Elon Musk says he doesn't care about benchmark scores, only wanting the model to be genuinely useful.

Beating AI Insight News Brief: Elon Musk acknowledged on October 30 that the development of Grok 4.8 took longer than expected, but guaranteed the new model "will not disappoint". He noted that the team does not optimize for benchmark tests, and he does not care about benchmark scores, focusing instead on the model's practical applications. On September 14, Musk revealed that Grok 4.8 has 2.5 trillion parameters, uses a new C++ training system, and was expected to complete pre-training by the end of that week before moving to reinforcement learning. More than six weeks later, the new model remains unreleased. Musk’s emphasis on not caring about benchmark scores stems from Grok 4.7’s underwhelming performance. Independent evaluation firm Artificial Analysis found that under the same xhigh inference setting, Grok 4.7’s composite intelligence index only rose by 2 points from Grok 4.6 (44 to 46), while average output tokens per query jumped from around 38,000 to 81,000. Users in the Cursor community also complained that Grok 4.7 consumes quotas faster with negligible real-world improvements. In an interview with China Central Television (CCTV) in September, Musk admitted Grok 4.7 is inferior to Claude Opus 5.5, and is not expected to catch up with top-tier models until 2027. In October, he further proposed that Grok Bot call external models including Claude based on task requirements, and plans to have the high-speed version of Grok 4.8 handle a large volume of simple requests.