SAN FRANCISCO — OpenAI released GPT-5 to the public on Tuesday, and the artificial intelligence research community is still processing what they are seeing. The new model demonstrates reasoning capabilities, scientific problem-solving, and creative output that even the company\'s own researchers describe as surprising, performing at or above the level of domain experts in fields ranging from molecular biology to contract law.
GPT-5 represents what many researchers are calling a genuine leap in AI capability. Photo: Unsplash
In benchmark testing released alongside the model, GPT-5 scored in the 98th percentile on the US Medical Licensing Exam, outperformed 94 percent of practicing attorneys on the bar exam, solved 87 percent of problems from the International Mathematical Olympiad, and wrote functional code in 42 different programming languages. On creative writing benchmarks, human evaluators rated GPT-5\'s output as indistinguishable from or superior to professional human writers in 71 percent of cases.
What Makes GPT-5 Different
Previous AI models excelled at pattern recognition and text generation but struggled with genuine multi-step reasoning — the kind of thinking that requires holding a complex problem in mind, breaking it into components, working through each component systematically, and integrating the results. GPT-5 appears to handle this kind of reasoning in a fundamentally different way, according to researchers who have spent time with the model since its release.
Dr. James Huang, an AI researcher at MIT who has been testing the model since Tuesday morning, told reporters that it is not just doing better at the same things — it is doing things that previous models genuinely could not do at all. He described prompting the model with an unsolved problem in his own research area and receiving an analysis that identified an approach he had not considered.
Researchers say GPT-5 demonstrates capabilities that go beyond previous AI models. Photo: Unsplash
Safety and Concerns
OpenAI said it spent eight months on safety testing before releasing GPT-5, working with external red teams and government agencies to identify potential risks. The company said it implemented new content filtering systems and usage monitoring that go significantly beyond what was in place for previous models. Still, AI safety researchers at organizations including the Center for AI Safety said they have serious concerns about releasing a model with these capabilities before the field has developed adequate tools for understanding what the model is actually doing internally.
Pricing and Access
GPT-5 is available through OpenAI\'s API and ChatGPT interface. The base model is available to all ChatGPT users, while the most capable version requires a premium subscription. Enterprise pricing is available for businesses that need high-volume access.
Leave a Reply