Grok 4.5 entered private beta on June 28, 2026, at SpaceX and Tesla. The AI model is built on a 1.5 trillion-parameter V9 foundation with training data from Cursor, an AI-powered code editor. Elon Musk announced the development but no public benchmark data exists yet.

According to internal evaluations at SpaceX and Tesla, Grok 4.5 is close to or exceeds Opus performance. Opus is Anthropic’s most capable model. If Grok 4.5 truly matches Opus, it would be a significant achievement for xAI, Musk’s AI company founded in 2023.

The Training Data Bet

Incorporating Cursor data is strategic. Cursor is widely used by developers. The platform captures programming patterns, debugging workflows, and coding contexts that most AI models don’t see at scale. If Grok 4.5 learns from Cursor’s dataset, it could develop stronger programming capabilities.

This approach diverges from competitors. OpenAI and Anthropic rely on public internet data and partnerships. Musk’s access to Cursor data gives xAI a unique training advantage for coding-specific tasks.

Private Beta Signals Confidence

Private beta at SpaceX and Tesla first means Musk wants internal feedback before public release. SpaceX uses AI for software and operations. Tesla uses it for vehicle systems and manufacturing. Real-world deployment in these settings will reveal whether Grok 4.5 is actually competitive.

Grok 4.5 matters because xAI is moving from hype to capability. Private beta at Musk’s companies suggests the model is ready for real work, not just marketing.