On June 28, 2026, local time, Elon Musk posted on the social platform X that his latest large language model, Grok4.5, has started beta testing internally at SpaceX and Tesla, and will gradually expand its rollout in the future.

According to the disclosure, Grok4.5 is built upon a V9 base model with 1.5 trillion parameters, and during the supplementary training phase, it specifically integrated data from the popular AI coding tool Cursor. Musk stated that early evaluations show its performance is close to or even exceeds the flagship model of Anthropic, Claude Opus. Currently, reinforcement learning is still being continuously optimized, and the supporting test benchmarks are also being improved.

QQ20260629-134843.jpg

What has attracted more attention is its release plan: Musk said that SpaceX will launch a new foundational model "trained completely from scratch" every month for the rest of this year, rather than a simple iteration of the existing version. This pace far exceeds the current industry standard cycle. If successfully implemented, it will further intensify competition with leading companies such as OpenAI and Google.

This move also corresponds to recent strategic actions — on the 16th of this month, SpaceX announced plans to acquire Anysphere, the developer of Cursor, for $6 billion, with the transaction expected to be completed by the third quarter of 2026. Testing it first in its own physical industry scenarios not only provides high-value feedback for model optimization but also highlights xAI's determination to deepen its presence in the enterprise-level AI market and strengthen its coding assistance sector.