SpaceXAI launches Grok 4.5: the first model trained with Cursor

What happened?

On July 9, 2026, SpaceXAI (formerly xAI) unveiled Grok 4.5, the first artificial intelligence model trained in direct collaboration with Cursor. According to Engadget and WWWhat's new, the model is available today in the Grok Build programming agent, across all Cursor plans, and on the SpaceXAI console. This launch is the first since the official rebrand from xAI to SpaceXAI, which occurred the previous day, and marks a milestone in integrating SpaceX's AI capabilities with the most popular development tool of the moment.

Grok 4.5 was trained on tens of thousands of NVIDIA GB300 GPUs, with a special emphasis on data processing and curation. SpaceXAI describes it as its "smartest model to date," designed for coding, agentic tasks, and knowledge work. The Cursor team called it "the most powerful we've built, and the first we didn't make solely for software engineering." This multidisciplinary approach suggests SpaceXAI aims to compete not only in the coding agent niche but in general productivity assistants.

Why is this important?

This launch marks several milestones. It is the first post-rebrand model from SpaceXAI, the first public demonstration of the partnership with Cursor (which SpaceX is acquiring for $60 billion), and comes at a time when the coding agent race is becoming the most competitive segment of the AI market. Competitors like Anthropic's Claude Code and OpenAI's Codex are already positioned, and Grok 4.5 seeks to differentiate itself through price and speed.

The pricing is especially relevant: $2 per million input tokens and $6 per million output tokens, placing it below models like Anthropic's Opus (which costs $15/M output tokens) and directly competing with Claude Sonnet and GPT-4o. SpaceXAI claims it is designed to respond faster than "flash" models at much lower costs. In a demo, Grok 4.5 generated an interactive solar system simulation with a single prompt, illustrating its ability to generate complex functional code. While impressive, such demos do not replace independent benchmark tests.

Additionally, the historical context is relevant: xAI was founded in 2023 by Elon Musk as a direct competitor to OpenAI. After merging with SpaceX in 2025, the company was renamed SpaceXAI, integrating hardware and software resources. This move echoes Google's acquisition of DeepMind in 2014, but with the difference that Cursor is not a research lab but a tool with millions of active users. The $60 billion acquisition of Cursor is one of the largest in the tech sector, surpassing Microsoft's purchase of GitHub ($7.5 billion in 2018) and reflecting the strategic importance of development tools.

Consequences and context

The alliance with Cursor is not trivial. Cursor is one of the most popular AI-assisted programming tools, with over 2 million active developers according to 2025 data. Its integration with Grok 4.5 could accelerate model adoption among developers, but SpaceX's $60 billion acquisition of Cursor raises questions about the tool's future independence and potential vertical integration. Historically, when a large company acquires a development tool, tensions arise: for example, Microsoft's acquisition of GitHub initially generated distrust, though neutrality was later maintained. In this case, the risk is that Cursor may prioritize SpaceXAI models over others, limiting user choice.

For users, Grok 4.5 offers a cheaper alternative to competing models, but whether its performance on real engineering benchmarks surpasses Claude Code or Codex remains to be seen. SpaceXAI claims it outperforms real engineering benchmarks, but detailed results have not been published. In the past, other companies have made similar claims that were later unconfirmed by independent tests (e.g., LLaMA 2 vs GPT-3.5 benchmarks). The lack of transparency is a weak point that SpaceXAI must address to gain credibility in the developer community.

The coding agent market is booming. According to Gartner, the market for AI-assisted development tools is expected to grow from $1.5 billion in 2025 to $8 billion in 2028. SpaceXAI's entry with a model specifically trained for programming, backed by Cursor's infrastructure, could reshape the landscape. However, speculation about future integration and the lack of independent benchmarks are points to watch. Compared to the launch of GitHub Copilot in 2021, which revolutionized the market, Grok 4.5 arrives in a more mature environment where differentiation by price and speed may be key.

Another aspect to consider is the impact on independent developers and startups. If Grok 4.5 offers performance comparable to more expensive models, it could democratize access to advanced coding assistants. However, vertical integration with Cursor could create a closed ecosystem, similar to Apple's with Xcode and its frameworks. This could harm competitors like Replit or CodeSandbox, which rely on open models.

What readers should know

• Grok 4.5 is available today in Grok Build, all Cursor plans, and the SpaceXAI console.
• Pricing: $2/M input tokens, $6/M output tokens, significantly cheaper than Opus ($15/M) and competitive with Sonnet ($3/M input, $15/M output).
• Trained on NVIDIA GB300 GPUs (tens of thousands) with an emphasis on data curation to improve generated code quality.
• It is SpaceXAI's first model after the rebrand from xAI, which occurred on July 8, 2026.
• SpaceX's acquisition of Cursor ($60 billion) is ongoing and expected to close in Q4 2026.
• SpaceXAI claims it outperforms real engineering benchmarks but has not published detailed results; independent tests like HumanEval or SWE-bench are recommended.
• The model is designed not only for coding but also for agentic tasks and knowledge work, positioning it as a general assistant.

"The Cursor team itself defined the model as 'the most powerful we've built, and the first we didn't make solely for software engineering.'"

— Statement reported by Engadget and WWWhat's new, July 9, 2026.