New ChatGPT Features – June 2025 (Bonus Feature)

There have been some big updates this week that are actually worth knowing about that I haven’t already covered in my larger roundup — especially if you’re using GPTs for building, teaching, collaborating, or just getting stuff done.

Here’s a quick breakdown of what’s new and why it matters.

ChatGPT Bonus Features June 2025 is prominently displayed with a stylized emblem above the text.

Custom GPTs: Use Any Model You Like (Almost)

Pro, Plus, and Team users can now choose from all the available models — when building or using Custom GPTs without Actions, meaning you are no longer limited to 4o.

You can also now set a recommended model when you publish a GPT, which is great if you’ve built something that works better with a specific one.

Personally, I’ve been using GPT-4.1-Mini more and more lately. It’s fast, sharp, and gets through jobs without making a fuss.

Worth noting: If your GPT uses Actions (APIs or external tools), you’re still a bit restricted in terms of model choice — but that’s expected to change soon.

Projects: Now a Proper Workspace

Projects are becoming much more useful — not just a container for chats, but something that can actually support real workflows.

You can now:

  • Run proper deep research directly inside a Project
  • Use Advanced Voice Mode with your Project
  • Take advantage of Project-specific memory
  • Share individual chats (finally)
  • Upload files from mobile, and pick which model you want to use
  • Instantly turn any chat into a new Project

Side note: You still can’t move a chat from CustomGPT into a project though, but here’s hoping 🤞

This brings a lot of small but important changes together. The ability to work across voice, files, and scoped memory makes Projects feel a lot more like a working environment and not just a conversation thread.

Advanced Voice Mode Is Now Even More Natural

Voice Mode got a stealth update and it’s much better now.

It’s not just clearer — it’s more human. There are pauses, thinking sounds, natural filler words… and it’s hard to tell it’s not a real person. If you haven’t tried it recently, it’s worth another go.

This could open up some interesting use cases for voice-led tools, training, and even accessibility.

In Summary

None of these changes scream “headline feature” on their own — but together, they open the door to much better ways of working with GPT.

  • You can now use the right model for the job
  • Projects are starting to behave more like proper tools
  • Voice is becoming something you could actually use day-to-day

If you’re already building with GPTs or helping others learn how to use them, these updates give you more control and more ways to experiment.

Let me know if you want a quick rundown of which models are best for which jobs — I’ve got a few notes on that.