OpenAI: GPT-5.2 Chat

GPT-5.2 Chat (AKA Instant) demonstrates exceptional performance as a fast, lightweight model optimized for interactive chat. It consistently ranks among the fastest models, achieving an Infinityth percentile across all speed benchmarks, and offers highly competitive pricing, also at the Infinityth percentile. Reliability is outstanding, with a 100% success rate across all benchmarks, indicating minimal technical failures. The model exhibits perfect accuracy in General Knowledge, making it the most accurate at its price point and speed. It also shows strong capabilities in Instruction Following (91% accuracy, 96th percentile) and Coding (96% accuracy, 97th percentile), with Instruction Following being the most accurate model at its price point. Mathematics performance is robust at 94% accuracy (84th percentile). While its Ethics and Email Classification scores are solid at 99% and 98% respectively, they fall within the 50th-55th percentile. A notable weakness is its 0.0% accuracy on the Hallucinations (Baseline) test, indicating it does not appropriately acknowledge uncertainty for fictional concepts. Reasoning is fair at 78% accuracy (64th percentile). Overall, GPT-5.2 Chat excels in speed, cost-efficiency, and reliability, making it ideal for high-throughput, interactive applications, though its handling of uncertainty requires attention.