OpenAI Releases GPT-5 with Multimodal Reasoning Across Text, Audio, and Video
The new model demonstrates significantly improved performance on scientific and legal reasoning benchmarks.
OpenAI on Thursday unveiled GPT-5, its most capable language model to date, featuring native multimodal reasoning across text, audio, and video inputs.
The model showed substantial improvements on academic and professional benchmarks. On the bar exam, GPT-5 scored in the 94th percentile compared to the 90th for its predecessor. On medical licensing exams, it matched or exceeded the performance of licensed practitioners in 11 of 14 specialty areas.
CEO Sam Altman described the release as a "step-change in useful intelligence," though he acknowledged that the model still makes factual errors and can be confidently wrong on obscure topics.
GPT-5 will be available to ChatGPT Plus subscribers immediately, with API access rolling out to developers over the following two weeks.
Competitors Google DeepMind and Anthropic both responded within hours of the announcement. Regulators in the EU said they would begin reviewing GPT-5's compliance with the EU AI Act within 30 days of public release.