OpenAI’s latest update lets ChatGPT users choose how fast GPT-5 thinks: How the new feature works

OpenAI has introduced a new feature that gives ChatGPT users more control over how quickly GPT-5 with Thinking responds. The update, announced on X on September 18, allows Plus, Pro and Business subscribers to choose the “thinking time” of the model directly from the message composer on web.

OpenAI adds speed control to GPT-5 Thinking

Previously, GPT-5 with Thinking often took longer to generate responses, as the model was designed to reason more carefully before answering. Many users, however, said they wanted more flexibility to balance depth and speed depending on the task.

Under the new system, subscribers can now toggle between different pacing modes:

Standard: The new default for all plans, offering a balance between speed and intelligence.

Extended: The former default for Plus users, where the model takes more time to produce deeper responses.

Light: Exclusive to Pro users, delivering the snappiest replies for quick tasks.

Heavy: Also exclusive to Pro, designed for the deepest reasoning, with slower but more comprehensive outputs.

OpenAI says the chosen mode will remain active across future chats on the web until users manually change it.

“We’ve heard your feedback that GPT-5 Thinking can sometimes take longer than you’d like. Now Plus, Pro, and Business users can set the pace to match the moment,” wrote the AI company on its X account.

GPT-5 Codex for developers

Meanwhile, OpenAI has also unveiled GPT‑5 Codex, a new version of its AI designed specifically for software development, earlier this week. The model is built to help developers write, debug, and review code more efficiently, working both interactively with humans and independently on complex tasks.

OpenAI claims that GPT‑5 Codex can handle a wide range of programming tasks, from creating projects from scratch to adding features, fixing bugs, and checking other people’s code. It can work on long projects by itself for hours, testing and improving its own code until it works properly. For smaller, simpler tasks, it responds quickly without wasting resources.

The AI has been trained to review code carefully and find critical errors before they cause problems, said OpenAI in its blog post. Tests on popular open-source projects showed that its suggestions are more accurate and useful than before, helping developers focus on the most important issues.