AI coding models are starting to compete on more than raw benchmark scores. With Grok 4.5, SpaceXAI is betting that lower token usage and better real-world coding workflows matter just as much as topping another leaderboard.
SpaceXAI released Grok 4.5 on July 8, 2026, calling it the company’s smartest model yet. It’s built around coding, agentic tasks, and professional knowledge work. SpaceXAI, formerly known as xAI before its July 6 rebrand following SpaceX’s acquisition of the company, trained the model in partnership with Cursor, the AI coding editor.
Coding Performance
Grok 4.5 was trained on coding, science, engineering, and math data. SpaceXAI says it exceeds comparable leading models on real engineering tasks. It also scores #1 on Harvey’s Legal Agent Benchmark, a test of office and knowledge work.
On coding benchmarks specifically, SpaceXAI’s own numbers put Grok 4.5 in the middle of the pack among top models, not at the very top. On SWE-Bench Pro it scores 64.7%. That’s behind Opus 4.8 at 69.2% and Anthropic’s Fable model at 80.4%. On Terminal-Bench 2.1, Grok 4.5 scores 83.3%, close behind GPT-5.5 and Fable.
Efficiency
Where Grok 4.5 does lead is efficiency. SpaceXAI says it runs at 80 tokens per second. It also uses about 4.2 times fewer output tokens than Opus 4.8 to solve the same SWE-Bench Pro tasks, 15,954 tokens on average versus 67,020. Grok 4.5 is now the default model in Grok Build, SpaceXAI’s coding tool. It can build complete Excel models with web research and multi-sheet formulas, and it works inside Word and PowerPoint using native formatting and shapes.
Availability and Pricing
Grok 4.5 costs $2 per million input tokens and $6 per million output tokens. It’s available today in Grok Build, in Cursor on all plans, and through the xAI API console. It is not yet available in the EU in any SpaceXAI product or API. EU access is expected in mid-July.
Why It Matters
The efficiency numbers are the real story here, more than the raw benchmark scores. Grok 4.5 doesn’t top every coding leaderboard. Using a fraction of the tokens to get close to models that do is a different kind of win. For anyone paying per token, a model that solves the same task in 16,000 tokens instead of 67,000 is doing meaningfully cheaper work, even if its ceiling on the hardest problems sits a bit lower.
The Cursor training partnership is also worth noting. SpaceXAI built Grok 4.5 together with a coding tool company, rather than building a model first and adapting developer tools afterward. That’s a sign of how tightly model development and the tools built around it are starting to merge.
Should You Care?
If you code with Cursor, or you’re choosing an API model for coding and agent work, yes. Grok 4.5 is worth testing directly against whatever you’re using now, especially if token cost has been a real factor in that decision. The free usage window in Grok Build and Cursor makes that an easy thing to try before committing to anything.
If you’re in the EU, there’s nothing to test yet. Grok 4.5 isn’t available there in any SpaceXAI product or through the xAI API console until access opens up in mid-July.
Source: SpaceXAI: Introducing Grok 4.5