Grok 4.5

Grok 4.5 debuts with strong performance across reasoning and coding tasks

Grok 4.5 enters the benchmark as xAI's latest flagship model, establishing a solid baseline across diverse evaluation categories. The model demonstrates particularly strong performance in mathematical reasoning, achieving 88.3% on MATH-500 and ranking competitively on AIME 2024. Coding capabilities are robust with 86.8% on HumanEval and 90.8% on a private coding evaluation. General knowledge and instruction following show respectable scores of 88.6% on MMLU-Pro and 79.4% on IFEval respectively. The model supports multimodal capabilities including vision, structured JSON output, tool use, and an extended 128K context window. Performance on vision tasks is moderate at 72.7% on MMMU, suggesting room for improvement in multimodal reasoning. Graduate-level science reasoning scores 61.6% on GPQA Diamond, placing it in the middle tier for this challenging benchmark. The model also includes reasoning-specific capabilities, though no dedicated reasoning benchmarks are provided in this window. As a first entry, Grok 4.5 establishes itself as a well-rounded model with particular strengths in mathematics and coding, while leaving opportunities for growth in vision understanding and advanced reasoning tasks.