# SpaceX's Grok 4.5: A 1.5 Trillion-Parameter MoE Model Trained on Cursor Data

> The newest Grok is a 1.5 trillion-parameter mixture-of-experts model, and its headline trick is not raw size but economics. It posts 83.3 percent on Terminal-Bench 2.1, a test of how well a model can drive a command line and finish real coding tasks, while burning roughly a quarter fewer output tokens than rival systems. Because you pay per token, that efficiency compounds: at $2 per million input tokens and $6 per million output, it undercuts older frontier models by anywhere from five to fifty times. Part of the edge appears to come from training on interaction data from Cursor, the AI code editor, which feeds the model a diet of real developer workflows rather than generic text. The result is a system tuned for the unglamorous grind of software work. The significance is the direction of travel. Frontier labs spent years competing on benchmark ceilings; the fight now is over cost per useful task, and a cheap, fast model that codes well is the kind of thing that quietly reshapes who can afford to build with AI.

_Section: [Daily AI Updates](https://www.wortins.com/daily-ai) · Source: KERSAI · Published Thursday, July 16, 2026_

## Wortins' read

The newest Grok is a 1.5 trillion-parameter mixture-of-experts model, and its headline trick is not raw size but economics. It posts 83.3 percent on Terminal-Bench 2.1, a test of how well a model can drive a command line and finish real coding tasks, while burning roughly a quarter fewer output tokens than rival systems. Because you pay per token, that efficiency compounds: at $2 per million input tokens and $6 per million output, it undercuts older frontier models by anywhere from five to fifty times. Part of the edge appears to come from training on interaction data from Cursor, the AI code editor, which feeds the model a diet of real developer workflows rather than generic text. The result is a system tuned for the unglamorous grind of software work. The significance is the direction of travel. Frontier labs spent years competing on benchmark ceilings; the fight now is over cost per useful task, and a cheap, fast model that codes well is the kind of thing that quietly reshapes who can afford to build with AI.

## Source

[Read the full story at KERSAI](https://kersai.com/ai-breakthroughs-july-2026/)

## Related coverage

- [OpenAI launches its new family of models with GPT-5.6](https://www.wortins.com/story/openai-launches-its-new-family-of-models-with-gpt-5-6-1f60cb54) — [TechCrunch](https://techcrunch.com/2026/07/09/openai-launches-its-new-family-of-models-with-gpt-5-6/)
- [Another contender just joined the arms race over AI in schools](https://www.wortins.com/story/another-contender-just-joined-the-arms-race-over-ai-in-schoo-39958f53) — [Chalkbeat](https://www.chalkbeat.org/2026/07/14/anthropic-launches-claude-for-teachers-as-ai-companies-battle-for-classrooms/)
- [Helsing Raises $1.8B Series E at $18B Valuation in Europe's Largest Defense-Tech Round](https://www.wortins.com/story/helsing-raises-1-8b-series-e-at-18b-valuation-in-europe-s-la-32ef801c) — [cnbc.com](https://www.cnbc.com/2026/07/13/helsing-fund-raise-defense-18-billion.html)
- [Unitree Robotics Files for $7B Shanghai IPO with GD01 Mecha Launch](https://www.wortins.com/story/unitree-robotics-files-for-7b-shanghai-ipo-with-gd01-mecha-l-6a7fe21f) — [thenextweb.com](https://thenextweb.com/news/unitree-gd01-mecha-humanoid-robot-ipo)
- [Meta just launched a new AI generator, Muse Image, and users are already pushing back over use of their photos](https://www.wortins.com/story/meta-just-launched-a-new-ai-generator-muse-image-and-users-a-f69ca1c9) — [TechCrunch](https://techcrunch.com/2026/07/07/meta-rolls-out-muse-a-new-ai-image-generator/)
- [Neocloud Together AI raises $800M, leaps to $8.3B valuation](https://www.wortins.com/story/neocloud-together-ai-raises-800m-leaps-to-8-3b-valuation-fd68cc47) — [TechCrunch](https://techcrunch.com/2026/07/01/neocloud-together-ai-raises-800m-leaps-to-8-3b-valuation/)

---

_Curated and written by [Wortins](https://www.wortins.com) — The daily AI briefing. Every story links to its original source; the "Wortins read" on each is our own original analysis. [About Wortins & our editorial approach](https://www.wortins.com/about)._