Grok 4.6 gets Aug. 7 launch date, Grok 4.7 follows

Elon Musk has set an early August target for Grok 4.6 and disclosed plans for a larger Grok 4.7 model as SpaceXAI steps up its competition with OpenAI, Anthropic, and China’s Moonshot AI.

Grok 4.6 targets an Aug. 7 release

Musk disclosed the timeline in a July 28 post on X, saying Grok 4.6 would be released “around August 7.” He described it as a 1.5 trillion-parameter model with stronger supervised fine-tuning and reinforcement learning.

Supervised fine-tuning trains a model on selected examples to improve the quality of its responses. Reinforcement learning uses feedback and reward signals to refine how a model handles tasks and follows instructions.

Musk also said Grok 4.7 would follow “a few weeks later” as a 2.1 trillion-parameter model. However, he did not provide an exact launch date or disclose pricing and availability details for either release.

“This will be better than 4.6 in every way, except slightly slower to serve, albeit with even better token efficiency,” Musk said about Grok 4.7.

Both dates remain targets rather than confirmed release appointments. Model development schedules can change during training, testing, and deployment.

Grok 4.5 sets the baseline for the upgrades

SpaceXAI introduced Grok 4.5 earlier in July as a model designed for coding, agent-based tasks, and knowledge work. The company currently charges $2 per million input tokens and $6 per million output tokens.

Musk announced the next two models in response to an assessment from Vercel CEO Guillermo Rauch, who described Grok 4.5 as the best-performing cybersecurity model relative to its price in Vercel’s latest tests.

According to Rauch’s post, Grok 4.5 was 10 times cheaper than OpenAI’s GPT-5.6 Sol, 5.7 times cheaper than Anthropic’s Opus 5, and 2.2 times cheaper than Moonshot’s Kimi K3 while delivering performance close to Kimi.

Rauch still ranked Sol as the leading frontier model, ahead of Opus 5. His findings reflect Vercel’s testing methods and do not establish a universal ranking across every AI workload.

SpaceXAI has not said whether Grok 4.6 and Grok 4.7 will retain Grok 4.5’s pricing. Service speed, token use, and API costs will determine whether the newer models maintain the same price-performance advantage.

US AI firms face pressure from China’s Kimi K3

Grok’s planned upgrades arrive as Chinese developers narrow the gap with leading US AI companies. Moonshot AI released the full weights for Kimi K3 on July 27, allowing developers to download, modify, and operate the model on their own infrastructure.

Kimi K3 has 2.8 trillion parameters, native visual capabilities, and a one-million-token context window. Its scale and open-weight structure have added pressure on US companies that mainly distribute their most advanced models through closed platforms and paid APIs.

SpaceXAI, OpenAI, and Anthropic are also competing on coding, cybersecurity, reasoning, and the cost required to complete each task. Grok 4.7’s larger architecture may improve its capabilities, but Musk acknowledged that it would be slower to serve than Grok 4.6.

Parameter counts alone do not determine model quality. Training data, architecture, post-training methods, inference systems, and token efficiency can all affect performance.

US AI firms split over federal oversight

OpenAI and Anthropic have already agreed to provide US officials with early access to unreleased frontier models for safety testing. Those agreements formed the basis for a broader voluntary federal review framework that now covers other major developers, including SpaceXAI, Google DeepMind, and Microsoft.

President Donald Trump formalized the framework through a June executive order. Participating companies can provide covered models to the federal government for up to 30 days before releasing them to trusted partners. The order states that the reviews do not create a mandatory licensing or preclearance system.

However, the industry remains divided over how Washington should handle open-weight models. Nvidia, Microsoft, Meta, OpenAI, Palantir, and other organizations warned US policymakers against broad restrictions in a July 24 open letter. They argued that open models support competition, lower deployment costs, and allow developers to run and modify AI systems on their own infrastructure.

The debate has intensified following Moonshot AI’s release of the 2.8 trillion-parameter Kimi K3. Trump administration officials have accused Moonshot of using outputs from Anthropic’s Fable model to train Kimi K3 through large-scale distillation. White House technology adviser Michael Kratsios described the alleged practice as an attempt to obtain proprietary US technology, while Treasury Secretary Scott Bessent said the administration was considering sanctions and placing Moonshot on a trade blacklist.

Moonshot has not publicly accepted those allegations. The dispute has nevertheless placed Kimi K3 at the center of a wider policy fight over whether Washington should target specific security and intellectual-property risks or restrict access to Chinese open-weight models more broadly. Grok 4.6 and Grok 4.7 will enter that market as US developers face pressure to improve performance while meeting emerging federal safety expectations.