OpenAI Releases GPT-5 with Native Multimodal Reasoning

OpenAI launched GPT-5, describing it as the company's first model trained end-to-end on text, audio, image, and video with a unified token space. In internal benchmarks, the model outperforms GPT-4o on MMLU-Pro and achieves near-human performance on the FrontierMath benchmark.

GPT-5 is available immediately to ChatGPT Pro subscribers and will roll out to Plus users over the next two weeks. Sam Altman said the model shows early signs of general reasoning, but the company also disclosed new safety mitigations for long-horizon planning behavior.