Elon Musk updates the Grok roadmap on X on 9/14: Grok 4.8 (about 2.5 trillion parameters, a new C++ training stack) is expected to finish training by the end of that week, then move into reinforcement learning (RL).
Also explained on the same day: The current 4.7 is broadly aligned with Anthropic Opus 5.0, but it isn’t 5.1 yet; multimodal capabilities still need to be improved. 4.8 will be a “noticeable improvement”; he estimates that 4.9 is in the Astra/Fable scale. As for Grok 5, he wrote “maybe better than anything. We shall see.” When asked how close 4.8 is to AGI, he only replied, “That will be Grok 5.”
This is the founder’s publicly stated schedule and self-assessment, not an independent evaluation report. For ordinary people: what you can watch in the short term is when 4.8 will truly enter RL, and whether there will be official releases/benchmarks afterward; as for Grok 5, it is still a direction, not a downloadable product commitment.
Source:
https://x.com/Stardocked/status/2100443714048974894
#xAI #Grok #Aİ
Also explained on the same day: The current 4.7 is broadly aligned with Anthropic Opus 5.0, but it isn’t 5.1 yet; multimodal capabilities still need to be improved. 4.8 will be a “noticeable improvement”; he estimates that 4.9 is in the Astra/Fable scale. As for Grok 5, he wrote “maybe better than anything. We shall see.” When asked how close 4.8 is to AGI, he only replied, “That will be Grok 5.”
This is the founder’s publicly stated schedule and self-assessment, not an independent evaluation report. For ordinary people: what you can watch in the short term is when 4.8 will truly enter RL, and whether there will be official releases/benchmarks afterward; as for Grok 5, it is still a direction, not a downloadable product commitment.
Source:
https://x.com/Stardocked/status/2100443714048974894
#xAI #Grok #Aİ