OpenAI released GPT-5 this week, and early benchmarks show significant improvements in reasoning, code generation, and mathematical problem-solving compared to its predecessor. The model scores 92% on the MMLU benchmark, up from 86%.

The new model introduces a chain-of-thought reasoning system that allows it to break down complex problems into steps, producing more accurate and explainable answers. It also features improved context handling with a 128,000-token window.

Industry experts said the improvements are most pronounced in STEM tasks, where GPT-5 now rivals specialized models. However, some noted that the model still struggles with certain types of hallucination and common-sense reasoning.

OpenAI said GPT-5 will be available through the API starting next month, with ChatGPT Plus users gaining access within weeks. The company did not disclose pricing for the new model.