Grok 4.5 Hits 1.5 Trillion Parameters — Cursor Partnership Cuts Token Cost to One-Quarter of Opus

On July 9, 2026, SpaceX AI released Grok 4.5, its first major model since absorbing xAI and rebranding. Unlike the older Grok versions, the new release is the result of a joint effort with Cursor, the popular AI coding editor that SpaceX acquired for roughly 60 billion dollars just weeks earlier. The model sits on a 1.5-trillion-parameter V9 base, three times larger than the previous V8 small model.

What makes the launch unusual is not raw size, but token efficiency. On the SWE-bench Pro coding benchmark, Grok 4.5 finishes each task using about 16,000 tokens, while Claude Opus 4.8 needs close to 67,000 for the same job. The model also runs at around 80 tokens per second, faster than most flagship models. Together, these two numbers mean that a developer can ship a complex agent workflow at a fraction of yesterday's cost.

The business case is built on three pillars. First, SpaceX gets Cursor's real developer interaction data, which captures how people actually use an editor, not just clean code. Second, Cursor customers can now call Grok 4.5 inside the same tool, with no extra setup. Third, the price is set at 2 dollars per million input tokens and 6 for output, less than a quarter of Opus's top tier.

For enterprise buyers, the message is simple. In a market where every vendor now offers a frontier model, the new battleground is unit cost per task. With Grok 4.5, the cost curve has bent again, and competitors will have to decide whether to cut their own prices or invest in other differentiators such as reliability, safety, or integration depth.