Last week, OpenAI rolled out GPT-5 on ChatGPT, replacing all existing models with GPT-5 and the new GPT-5 Thinking mode. This upgrade was designed to seamlessly shift between capabilities depending on the task, offering a noticeable performance boost. Thinking mode, in particular, provides a deeper analytical process by revealing the steps the AI takes to form its answers. But the rollout hasn’t been without turbulence, prompting quick revisions and a flood of user feedback.

In response to vocal protests from GPT-4o enthusiasts who preferred its “personality,” OpenAI restored GPT-4o as a legacy model for paid subscribers, alongside several other older versions. As of now, users have eight different models to choose from, but GPT-5 remains the default. If you enjoy exploring new tools, try GPT-5 Thinking, just be ready for changes, as OpenAI’s lineup is likely to keep evolving.

AI agents are no longer a futuristic concept, they’re here. With OpenAI rolling out “agent mode” to all ChatGPT subscribers, these tools can now take a single prompt and autonomously plan and execute complex tasks, from scheduling meetings to drafting follow‑ups.

It’s a big leap forward, with exciting potential and real risks. Here’s what you need to know before you dive in.

Since March 2025, ChatGPT Plus subscribers have been given limited access to GPT‑4.5, a new model described as “good for writing and exploring ideas.” Early users have noticed a clear improvement: GPT‑4.5 feels more natural, nuanced, and emotionally intelligent, making it an excellent tool for brainstorming and creative projects.

The catch? Access is capped at a small number of messages each week (OpenAI hasn’t said how many), so users are strategic in how they use them. Think of it as a temporary upgrade on a flight, more legroom and better snacks, while you let the model stretch its wings on your biggest ideas.