OpenAI Launches GPT-5 with Lower Costs and Controversial Charts
OpenAI releases GPT-5 with a new routing system, massive context windows, and an updated API for agentic applications. The launch faces immediate scrutiny due to misleading benchmark charts that visually contradict the company's own data.
OpenAI releases GPT-5 to ChatGPT users and API developers, introducing a router that decides how long the model needs to "think" before responding. The new lineup features a 400K token context window with up to 128K output tokens, alongside pricing structures designed specifically for production environments. By making the model immediately available to everyone, OpenAI locks in massive network effects and commoditizes access to its latest technology.
The launch faces immediate scrutiny after OpenAI presents benchmark charts that visually contradict their own numerical data. In one example, a chart labeled "coding deception" shows GPT-5 at 50 percent with a shorter bar than a competing model at 47.4 percent, a figure later corrected to 16.5 percent. Other slides display values of 69.1 percent and 30.8 percent as bars of equal height, flipping the intended ranking and misleading viewers about the actual performance improvements.
On the technical side, OpenAI consolidates its API surface around the Responses API to serve as the primary building block for agentic applications. This updated interface allows developers to access image generation, Code Interpreter, file search, and remote Model Context Protocol servers from a single request. Benchmark tests show the new ChatGPT agent achieves top scores on ML-engineering tasks and full-stack coding workloads, indicating the model excels at both code-centric debugging and long-horizon, multi-skill projects.