Elon Musk said on Oct. 30 that development of Grok 4.8 has taken longer than expected, though he said the new model would not disappoint. He added that the team is not optimizing for benchmark tests and that he cares more about real-world usefulness than scores. Musk had previously said on Sept. 14 that Grok 4.8 has 2.5 trillion parameters and uses a new C++ training system, with pretraining expected to finish that week before moving into reinforcement learning. More than six weeks later, the model has still not been formally released. Earlier observations from Artificial Analysis showed that under the same xhigh reasoning setting, Grok 4.7’s intelligence index rose only from 44 to 46 compared with Grok 4.6, while average output per question increased from about 38,000 tokens to 81,000. Users in the Cursor community also complained that Grok 4.7 burns through usage limits faster without a clear improvement in performance.
Elon Musk said on Oct. 30 that development of Grok 4.8 has taken longer than expected, but said the new model would not disappoint.
Musk said the team is not optimizing for benchmark tests and that he does not care about benchmark scores, placing more weight on practical use instead.
On Sept. 14, Musk had said Grok 4.8 has 2.5 trillion parameters and uses a new C++ training system. At the time, he expected pretraining to finish that week, followed by reinforcement learning. More than six weeks later, the model still has not been formally released.
Independent evaluation firm Artificial Analysis previously found that under the same xhigh reasoning setting, Grok 4.7’s overall intelligence index was only 2 points higher than Grok 4.6, rising from 44 to 46. Its average output per question, however, increased from about 38,000 tokens to 81,000 tokens.
Users in the Cursor community also complained that Grok 4.7 consumes usage quotas faster, while the actual improvement is not obvious.
Musk had also said in a September interview with China Media Group that Grok 4.7 was behind Claude Opus 5.5, and that it might not catch up with the top tier until 2027. In October, he also proposed letting Grok Bot call external models, including Claude, based on the task, while planning for a high-speed version of Grok 4.8 to handle a large number of simple requests.
This article was originally published by Bit.Fan. For more cryptocurrency news and market insights, visit www.bit.fan. Disclaimer:
The market information, project data, and third-party content displayed on this platform are for industry information sharing only and do not constitute any form of investment advice or return commitment.
Cryptocurrency trading carries high risks. Users should fully assess their risk tolerance and make independent decisions. All profits, losses, and legal responsibilities are borne by the users themselves.
