OpenAI has officially launched GPT-5, its most advanced language model to date, featuring what the company calls "real-time reasoning" — the ability to think through complex problems step-by-step before generating responses.
What's New
The key breakthrough is GPT-5's enhanced chain-of-thought system. Unlike previous models that generated tokens sequentially, GPT-5 can pause, reconsider, and even correct itself mid-generation. Internal benchmarks show it solves complex math problems 40% more accurately than GPT-4 Turbo.
"We're moving from pattern matching to actual reasoning," said Sam Altman during the livestream announcement. "GPT-5 doesn't just predict the next word — it understands the problem."
Performance Gains
- Coding: 35% improvement on SWE-bench (real-world GitHub issues)
- Math: 40% better on competition-level problems
- Science: Near-human performance on PhD-level physics questions
- Multilingual: Supports 50+ languages with improved accuracy
The model also introduces a new "thinking budget" parameter, allowing developers to control how much computation goes into each response — useful for balancing speed and accuracy.
Pricing and Availability
GPT-5 is available immediately through the API at $15 per million input tokens and $60 per million output tokens — a significant premium over GPT-4 Turbo. Free-tier users get limited access through ChatGPT.
The API rollout begins today for existing enterprise customers, with broader access coming in September.
Impact
This release intensifies the AI race with Anthropic, Google, and Meta. The "thinking budget" feature could change how developers architect AI applications — shifting from "prompt engineering" to "computation allocation."
OpenAI says GPT-5 will power its upcoming autonomous agent product, which can browse the web, write code, and complete multi-hour tasks independently.
Top comments (0)