Key facts
- SpaceXAI plans to release Grok 4.6 around August 7.
- Grok 4.6 will be a 1.5T parameter model with enhanced Supervised Fine-Tuning and Reinforcement Learning.
- Grok 4.7, a 2.1T parameter model, is slated for release a few weeks after Grok 4.6.
- Grok 4.5 was identified as the top cybersecurity AI model based on price-performance.
- Grok 4.5 is significantly more cost-effective than models like ChatGPT's Sol, Opus 5, and Kimi K3.
Elon Musk announced that SpaceXAI is preparing to launch its Grok 4.6 AI model around August 7, with a subsequent release of Grok 4.7 expected weeks later. The Grok 4.6 model will feature 1.5 trillion parameters and enhanced Supervised Fine-Tuning (SFT) and Reinforcement Learning (RL) capabilities.
Grok 4.7 is slated to be a 2.1 trillion parameter model, described by Musk as superior to Grok 4.6 in all aspects except for slightly slower serving times, while offering improved token efficiency.
These upcoming releases occur amid an intensifying AI race, with U.S. companies like SpaceXAI, OpenAI, and Anthropic facing competition from Chinese firms such as Moonshot, which recently released its open-weight Kimi K3 model. Reports suggest the Trump administration is considering restrictions on Chinese AI models to protect U.S. leadership, a move opposed by tech firms including NVIDIA, Elon Musk, and OpenAI CEO Sam Altman.
Separately, Vercel CEO Guillermo Rauch assessed Grok 4.5 as the leading cybersecurity AI model in terms of price-performance, noting it is significantly cheaper than competitors like ChatGPT's Sol, Opus 5, and Kimi K3 while delivering comparable performance. Meanwhile, Anthropic and OpenAI have jointly called for a federal review framework for advanced AI models to establish consistent safety standards for U.S. firms.