Model release · July 8, 2026
Grok 4.5
An Opus-class model, built from the ground up for code.
xAI's newest release trades general-purpose polish for something sharper: a model trained specifically for coding and long-running agentic work — cheaper to run, faster to answer, and tuned to operate across whole codebases with minimal hand-holding.
What xAI shipped
Grok 4.5 is built on xAI's 1.5-trillion-parameter V9 foundation model and was trained in partnership with Cursor to handle jobs that span multiple repositories and hundreds of tool calls.
Where it lands
Software-engineering and agentic scores from xAI's launch materials. Gold marks a category-leading result among evaluated models.
On tau3-Banking — described as the hardest financial-reasoning sub-task — Grok 4.5's 33% edged out GPT-5.5 (xhigh) at 31%, the top result among models xAI evaluated.
Same job, a quarter of the tokens
On SWE-Bench Pro, xAI reports Grok 4.5 resolves a task in far fewer output tokens than the field — the lever behind its lower running cost.
≈ 4.2× fewer output tokens per resolved task
An Opus-class model — at a fraction of the tokens, and a fraction of the cost.
— Elon Musk, describing Grok 4.5 · as reported by TechCrunch, July 8, 2026
Where to run it today
Read xAI's full Grok 4.5 announcement
Benchmarks, methodology, and rollout details, straight from xAI.
Open the announcement on x.ai →