Below the hood, there’s a “good router” that decides on the fly whether or not to make use of the usual GPT-5 or the extra superior GPT-5 Considering mannequin based mostly in your query’s complexity, device wants, or in case you explicitly say one thing like “assume arduous about this.” That router is continually enhancing by studying from actual utilization information. And in case you hit your GPT-5 restrict, you’ll robotically drop into GPT-5 Mini to maintain going.
OpenAI says it has made main strides in decreasing hallucinations, enhancing how the mannequin follows directions, and reducing down on the “tell-you-what-you-want-to-hear” drawback – all whereas dialing up efficiency in three key areas: writing, coding and well being.
Yep, well being. OpenAI claims GPT-5 is now higher at flagging potential points, tailoring solutions to your information degree and placement, and suggesting good follow-up questions on your physician. Nonetheless, you most likely don’t desire it making your medical calls.
GPT-5 is the primary time that it actually feels such as you’re speaking to an knowledgeable in any subject. It jogs my memory of when the iPhone went from these big, outdated pixel [screens] to the Retina Show, after which I went again to utilizing a kind of huge pixelated issues and I used to be like, ‘Wow, I am unable to imagine how unhealthy we had it.’
– Sam Altman, OpenAI CEO, August 7, 2025
One other huge shift? GPT-5 is now OpenAI’s default reasoning mannequin. Earlier than, you needed to manually choose “Assume Longer” to faucet into superior reasoning, that means a variety of free customers by no means knew it existed. Now it is simply in-built.
You should use GPT-5 to construct web sites and even create apps, or faucet into its writing capabilities to assist with on a regular basis duties like emails and modifying. | Video credit score – OpenAI
There are solely three flavors of GPT-5: the usual mannequin, GPT-5 Mini (for whenever you hit your restrict) and GPT-5 Professional (unique to the $200/month Professional plan). Plus customers get “considerably” extra utilization than free accounts, whereas Professional customers can go limitless.
The replace additionally brings some good extras to ChatGPT:
- Chat colour themes – with some unique shades for paying prospects.
- Simpler Google integration – ChatGPT can now robotically faucet into Gmail, Google Calendar, and Contacts with out you having to manually hyperlink them each time. Professional subscribers get it first subsequent week, with others following.
- Higher voice mode – OpenAI’s Superior Voice function is now smarter at understanding tone and adapting on the fly. The older Commonplace Voice Mode is being retired, liberating up assets for a smoother, extra pure voice expertise.
And naturally, there’s the security angle. OpenAI says GPT-5 strikes past the outdated “refuse or comply” coaching to one thing referred to as secure completions – the place the mannequin tries to provide the most useful reply doable whereas staying inside security traces. That may imply answering solely at a excessive degree, explaining why it might probably’t go additional, or providing secure alternate options. OpenAI says this method handles difficult, dual-use subjects extra gracefully and avoids pointless refusals with out opening the door to misuse.
And misuse is one thing that we already see taking place with unhealthy actors utilizing AI for misinformation and scams. So, tech firms ought to certainly assume extra about security (Galaxy telephones, for instance, would possibly quickly warn you about voice phishing) if they’re giving us all such highly effective instruments.
Talking of highly effective instruments, with Google and OpenAI locked in a head-to-head race, do not be shocked if we see a giant Gemini AI replace from Google any day now.
