OpenAI: GPT-4o Search Preview

OpenAI's GPT-4o Search Preview, created on March 12, 2025, is a specialized model designed for web search within Chat Completions, excelling at understanding and executing web search queries. This model consistently ranks among the fastest available, demonstrating exceptional speed across all benchmarks. Furthermore, it offers highly competitive pricing, consistently placing in the Infinityth percentile for cost-efficiency. Reliability is a standout feature, with the model achieving 100th percentile performance, indicating minimal technical failures and consistent, usable responses. In terms of benchmark performance, GPT-4o Search Preview shows strong capabilities across various domains. It achieved an impressive 89.0% accuracy in Coding (Baseline), coupled with a low cost and moderate duration. Its Reasoning (Baseline) performance was also robust at 88.0% accuracy, again with competitive pricing. The model demonstrated near-perfect accuracy in Ethics (Baseline) at 99.0% and strong General Knowledge (Baseline) at 96.0%, both with very low costs. A notable weakness is its 0.0% accuracy in Instruction Following (Baseline), suggesting this is an area requiring significant improvement. Email Classification (Baseline) showed solid accuracy at 96.0% with good cost efficiency. Overall, its strengths lie in ethical reasoning, general knowledge, and core coding/reasoning tasks, while instruction following presents a clear limitation.