Search Results: GPT5

OpenAI announces significant safety improvements to GPT-5, reporting a 65% reduction in harmful responses during mental health conversations. The updates follow lawsuits and expert criticism about AI's dangerous therapeutic limitations, even as CEO Sam Altman sends mixed signals about ChatGPT's role in emotional support.

Amp has permanently integrated GPT-5 as its new oracle, replacing the o3 model, to tackle complex development tasks like code refactoring and bug diagnosis. While Sonnet remains the primary agent for its proactive problem-solving, GPT-5's superior reasoning and planning capabilities offer a powerful complement. This strategic shift highlights the evolving role of specialized AI in enhancing developer workflows.

OpenAI is testing free trials for its $20/month ChatGPT Plus tier while expanding its budget-friendly $4 GPT Go plan to Indonesia, signaling a strategic push to broaden generative AI adoption. This tiered approach lowers entry barriers for developers and businesses, offering scaled access to GPT-5 and advanced features. The moves highlight OpenAI's focus on monetization and global accessibility in a competitive AI landscape.

OpenAI is extending its $4/month GPT Go subscription beyond India to new markets like Indonesia, with US expansion hinted by USD pricing. The budget tier offers access to GPT-5, enhanced image generation, and custom GPTs—positioning it as an affordable gateway to advanced AI capabilities.

OpenAI is recruiting 'AI-pilled' academics to build a scientific discovery accelerator platform leveraging GPT-5. This ambitious initiative aims to transform how research is conducted by automating hypothesis generation and experimental design, though skepticism remains about the model's capabilities.

Developer tests GPT-5's nano, mini and full models using an unusual benchmark – generating SVG pelicans riding bicycles – revealing surprising cost-performance tradeoffs. The $0.0054 mini model outperformed nano in speed while the premium $0.04 full version delivered cleaner outputs, challenging assumptions about model efficiency.