xAI Launches Grok 4.5: Coding and Agentic AI Model
Original: Grok 4.5
Why This Matters
Grok 4.5 enters a competitive frontier model market with claimed efficiency and speed advantages relevant to enterprise coding workloads.
xAI launched Grok 4.5 on July 8, 2026, describing it as its strongest model to date. Built for coding, agentic tasks, and knowledge work, it was trained on tens of thousands of NVIDIA GB300 GPUs and co-developed alongside Cursor.
xAI released Grok 4.5 on July 8, 2026, positioning it as the company's most capable model yet, optimized for software engineering, agentic workflows, and technical knowledge work. The model was trained in collaboration with Cursor and ran across tens of thousands of NVIDIA GB300 GPUs.
On DeepSWE 1.0, Grok 4.5 scored 62.0%, placing third behind Fable (max) at 66.1% and GPT 5.5 (xhigh) at 64.31%, and ahead of Opus 4.8 (max) at 55.75%. On Terminal Bench 2.1, Grok 4.5 scored 83.3%, nearly matching GPT 5.5 at 83.4%. On SWE Bench Pro, it achieved a 64.7% resolve rate.
xAI reports that Grok 4.5 is served at 80 tokens per second and delivers twice the token efficiency of comparable leading models on the same tasks, reducing output costs. Training employed large-scale reinforcement learning covering hundreds of thousands of tasks, with a focus on multi-step software engineering and automated and model-based grading. Data preparation included deduplication, quality scoring, and domain-focused selection. The model is available via the xAI API and developer console.