In a highly anticipated announcement, OpenAI has launched its new flagship artificial intelligence model, GPT-5. Codenamed 'Orion', the model represents a quantum leap in artificial intelligence, moving beyond simple next-token prediction to advanced, multi-step logical reasoning and native agentic autonomy.
Native Multimodality: Vision, Audio, and Video
Unlike previous models that patch different modalities together, GPT-5 is trained natively on text, image, audio, and video streams. This allows it to understand temporal relationships in video and audio seamlessly. For example, GPT-5 can watch a live football match or a complex machinery tutorial and provide real-time commentary, diagnose mechanical faults, or answer highly specific questions about physical dynamics.
Proactive AI Agents
The defining feature of GPT-5 is its ability to operate as a proactive agent. Users can assign complex, long-running objectives, such as "research, build, and deploy a responsive news sitemap generator for a website." GPT-5 can plan the project, write and run code locally, debug compile errors, fetch APIs, and deliver a completed product without requiring constant human intervention. It features an advanced safety sandbox that restricts harmful actions while maximizing task success rates.
Drastic Reductions in Hallucinations
Through a novel training paradigm called 'Self-Correction and Reflection', GPT-5 reviews its own thoughts before outputting text. During testing, this led to a 90% reduction in factual errors and logical slip-ups compared to GPT-4. The model excels in mathematics, scientific coding, and legal analysis, scoring in the 99th percentile on professional bar examinations and coding olympiads.
Pricing, Access, and Developer APIs
GPT-5 is available immediately to ChatGPT Plus subscribers and enterprise clients. The developer API features a context window of 1 million tokens, with input costs reduced by 40% compared to previous frontier models. OpenAI also announced partnerships with major hardware vendors to integrate GPT-5 locally on next-generation laptops and smartphones for offline, low-latency processing.

