OpenAI: Codex Mini

OpenAI: Codex Mini, a fine-tuned version of o4-mini for Codex CLI, demonstrates moderate speed performance, ranking in the 37th percentile across benchmarks. Its pricing tends to be at premium levels, placing it in the 9th percentile. However, the model exhibits exceptional reliability with a 100% success rate across all evaluated benchmarks, indicating minimal technical failures. In terms of specific performance, Codex Mini achieves perfect accuracy in both Ethics and Reasoning benchmarks, notably being the most accurate model at its price point and among models of comparable speed in these categories. It also shows strong performance in General Knowledge (99.5% accuracy, 74th percentile) and Coding (95.0% accuracy, 95th percentile). A significant strength is its Instruction Following, where it ranks in the 91st percentile for accuracy. The primary weakness identified is in the Hallucinations Baseline, where it scores only 54.0% accuracy, suggesting a tendency to provide answers to fictional concepts rather than acknowledging uncertainty. Email Classification performance is solid at 98.0% accuracy.