SpaceXAI and Cursor Release Grok 4.5 at $2/$6 Per Million Tokens

SpaceXAI—the rebranded AI unit formed when Elon Musk's xAI merged with SpaceX—launched Grok 4.5 on July 8, 2026, in partnership with AI coding platform Cursor. Elon Musk described it as 'Opus-class, but faster, more token-efficient and lower cost.'

Key Points:

• Grok 4.5 is the first commercial model trained jointly with Cursor, which SpaceX acquired in June for $60 billion—giving it access to trillions of tokens of real developer-agent interactions.

• Pricing: $2 per million input tokens, $6 per million output tokens—dramatically undercutting Anthropic's Claude Opus 4.8 ($5/$25) and comparable to Claude Sonnet 5.

• Grok 4.5 uses 60% fewer output tokens on average to complete complex software engineering tasks compared to Claude Opus 4.8, according to independent evaluations.

• Benchmark scores sit slightly below Fable 5 and GPT-5.5 on raw reasoning metrics, but Grok 4.5 outperforms on real-world software engineering tasks (SWE Marathon: 29% vs. Fable 5's 24%).

• Currently unavailable in the EU due to ongoing compliance audits; tentative EU release expected late July.

• Available in Grok Build, in Cursor on all plans with doubled free usage limits, and via the SpaceXAI console API.

For software development teams, Grok 4.5 inside Cursor with doubled free usage limits is worth evaluating as a cost-efficient alternative for code generation. SpaceXAI's vertical integration—from orbital data centers to model training to IDE integration—creates a uniquely closed, fast-moving competitive ecosystem.

Why It Matters: At $2/$6 per million tokens with 60% token efficiency gains, Grok 4.5 represents the most aggressive price-performance entry in the frontier model market—forcing competitors to justify their pricing.