Refresh
OpenAI is moving through this livestream pretty quickly. Now, Greg Brockman, President of OpenAI, is on stage, and the focus is shifting back to developers and coding. He says that GPT-5 is the best model at agentic coding tasks.
Sam Altman is now back onstage to start the discussion around ChatGPT-5 and how it handles the Health prompt. We now have two ChatGPT users on stage, including one, Carolina, who was diagnosed with three forms of cancer in one week.
After receiving an email confirming the diagnosis, this user took a screenshot and uploaded it to ChatGPT and got an understandable explanation. Altman says that Health questions are one of the most common use cases for ChatGPT as well. A takeaway from both of them is that ChatGPT can be used to help learn a bit more and ‘regain some agency.’
Both of these users have used GPT-5 and note that it is a much faster, « almost a little alarmingly at first. » Another key impression, though, is that GPT-5 is more of a thought partner than just delivering an answer based on the prompt. They also put the biopsy results back in, and GPT-5 asked follow-ups to get a bit more detail and deliver a more nuanced response.
Of course, ChatGPT or GPT-5 is not a medical doctor, and you should still see one for diagnosis and emergencies.
OpenAI is also focused on improving safety and deception; in this case, GPT-5 will aim to maximize helpfulness within the allowed safety constraints. This will be in contrast to how previous models, like o3, might have refused on the grounds of safety before.
GPT-5 could answer the same prompt, by explaining why it can’t offer help and point you towards a third-party or even a manual. The demo shown off by OpenAI here is a prompt asking for how to work with ‘pyrogen.’
ChatGPT will be able to connect with your Google account
We’re now switching gears to ChatGPT’s Memory function. Pro, Plus, and Team subscribers will be able to connect ChatGPT to a Google Account for calendar access. In fact, it can connect with Gmail and Google Calendar, allowing GPT-5 to pull in your schedule and get some more context.
It can even remind you to respond to an email that you’ve received, read, but haven’t yet responded to. This will roll out first to Pro users next week, then Plus and Team at some point in the future.
If you’ve been wanting to change the color of your ChatGPT interface, OpenAI is making that possible – to a degree – for paid users. Even more impressive, though, is the ability to customize the personality of GPT-5 a bit for research use cases.
Voice mode is getting an upgrade
There is a lot of focus on coding with GPT-5, but we’ve also seen improvements in writing thus far. The teams at OpenAI have also been focused on further improving Voice and video functionality.
With GPT-5, voice mode is coming to all users, and in this demo, OpenAI is showing that you can be specific with how you want an answer. In this case, they demonstrated ChatGPT Voice’s ability to respond with just one word, summarizing Pride and Prejudice as ‘relationships.’
The biggest improvment, though, is that it sounds a lot more natural than even the current version in the app.
Now we’re in the midst of another GPT-5 demo that’s centered around its ability to write code, spanning over 200 lines in just a few minutes of this demo. It actually created a site, with many visual elements and even audio ones, to help one learn French with a partner.
And like that, GPT-5 completed the prompt and coded the demo, which the OpenAI team is now testing during the live-stream. Pretty, pretty fast.
For the next demo, OpenAI is pitting GPT-4o and GPT-5 against each other in writing a eulogy for its previous models. Here’s a look at what GPT-4o produced.
GPT-5 on the other hand has more of an understanding of tone and nuances. It sounds a bit more like a person with emotions wrote it … less dry.
Our first demo of GPT-5
We’re now getting our first demo of GPT-5, first with the model being asked to explain the Bernoulli effect, and it responded pretty quickly. The follow-up, though, is asking GPT-5 to create a visual, and OpenAI is explaining that it will take a bit longer to respond, and in that time also suggest the best course of action.
You can also ask GPT-5 to ‘think harder’ or ‘more precisely’ in the actual prompt. Even neater, you can see GPT-5’s thought process while it’s answering the prompt.
GPT-5 will start rolling out to all users today
Here’s some good news – OpenAI is rolling out GPT-5 to all users, with that process beginning today, including the Free tier as well as Plus and Pro subscribers. It won’t be a free-for-all with GPT-5 in the free tier, but you’ll be able to use it until allotments are hit. Plus and Pro subscribers will expectedly get higher rate limits.
A key area of focus with GPT-5 was improving reliability and being accurate with facts. OpenAI is also saying that it has specifically focused on reliability in health.
Now, we’re getting a closer look at GPT-5 from more team members at OpenAI. Here’s the kicker, also, OpenAI says that GPT-5 will even be available for free users.
As for what we’ll see during this livestream, we’ll be getting plenty of demos but are starting first with evaluations. OpenAI has just announced that GPT-5 has set a new level on several benchmarks, including SWE-Bench – it’s not the full story, but it’s a good tell of the performance here.
Sam Altman announces GPT-5
And here we go, OpenAI CEO Sam Altman is walking onto a stage, and that’s certainly a different setting than previous livestreams.
Altman is wasting no time, saying that they’re ushering in GPT-5 and it’s another important step to AGI. He believes that folks will enjoy using GPT-5 a lot more. Why? Well, the early sell here, as Altman puts it, is that chatting or talking with GPT-5 is like talking to an expert, and it can even write software on demand.
2 minutes to go
The visuals have changed again, and we’re now counting down from three minutes until the start. The music is also getting a bit more exciting with more sounds joining in.
The livestream has started
OpenAI has officially kicked off the livestream with under 10 minutes to go until 1PM ET/ 6PM BST. Right now, we have some sonically soothing sounds over « OpenAI Summer Update » flashing with colors and then spinning in circles on the screen. No mention of GPT-5 though.
While we wait for the livestream to kick off on OpenAI’s YouTube channel – and in the video above – the description of the video does give us a bit more detail. We’ve already known, thanks to a post on X from Sam Altman, that this will be a longer livestream, and there will be many, many presenters.
Here’s what the description reads: « Join Sam Altman, Greg Brockman, Sebastien Bubeck, Mark Chen, Yann Dubois, Brian Fioca, Adi Ganesh, Oliver Godement, Saachi Jain, Christina Kaplan, Tina Kim, Elaine Ya Le, Felipe Millon, Michelle Pokrass, Jakub Pachocki, Max Schwarzer, Rennie Song, Ruochen Wang as they introduce and demo GPT-5. »
Further, if you had any doubt that we’d be seeing GPT-5 today, this confirms it will be unveiled and demoed.
I just checked my ChatGPT Plus account one more time before this event kicks off, and still no sign of GPT-5. Just a long list of models, and as my colleague John-Anthony already wrote here, it could use a bit of a cleanup and streamlining.
I’m particularly excited about GPT-5’s rumored capability of automatically selecting the right model to handle the prompt or whatever the user has asked. That could speed things up before the AI even gets to work.
Less than 30 minutes until kick off
We’re just under 30 minutes until OpenAI tells us everything about GPT-5 and the livestream for the event is now appearing on YouTube. Though it’s just a countdown and a still teaser image showing « GPT-5 » and « OpenAI » in front of a colorful array of gradients.
You can see the livestream embedded above, and once it starts, you can watch it right from there.
One. Hour. Countdown
One hour to go until OpenAI reveals GPT-5 and we enter into the next generation of ChatGPT.
OpenAI still hasn’t got a livestream listing on YouTube yet but stay tuned to this liveblog and we’ll update with a link as soon as we have access.
One last look before everything changes – again. pic.twitter.com/P4y0iUvjihAugust 7, 2025
I’ve been moaning about the OpenAI model names for what feels like an eternity at this point, so I may as well add one last criticism before we move into the GPT-5 era.
My colleague Lance Ulanoff has shared an image of the ChatGPT model selection page as it stands right now, and quite honestly, it might just be the most confusing naming scheme on the planet.
In just over an hour, OpenAI will announce GPT-5 alongside other models that we believe to be called GPT-5-mini, GPT-5-nano, and GPT-5-chat. Hopefully, the company decides to add these new models and replace the likes of GPT-4o, o3, o4-mini, o4-mini-high, otherwise, the list in that image is going to get even longer and even more confusing.
Another ChatGPT livestream
It’s only been two weeks since OpenAI revealed ChatGPT Agent in a livestream, and yet here we are waiting for what many believe to be the next generation of AI model.
Is GPT-5 really going to be as impressive as its next generation name would suggest? Or are we going to see small performance improvements in line with an iterative upgrade?
Either way, OpenAI needs to showcase why users should care about GPT-5 in this upcoming livestream, and so far the company’s livestreams have been anything but engaging.
With less than 90 minutes until the reveal, we’re all getting excited here at TechRadar. After all this could be the ChatGPT upgrade that pushes AI even further into the future.
Two hours to go
Only two hours until the GPT-5 reveal, and we still don’t have a livestream link on YouTube to share.
As soon as OpenAI adds a listing for the livestream, we’ll be sure to add it to this live blog, so stay tuned as we delve into the future of AI (and ChatGPT).
Please fix the naming scheme
OpenAI’s AI model naming scheme is confusing. There’s 4o, o3, o3-mini, o3-high, just to name a few.
In a few hours, at least four more models will be added to the mix, and it begs the question: At what point will OpenAI’s marketing step up and make ChatGPT more user-friendly?
If the company chooses to add another subscription tier with the aforementioned ChatGPT Go, not only will we have an AI model naming issue, but users will start to get confused by subscription tiers, too.
Please, OpenAI, fix your naming scheme.
A new ChatGPT subscription
New images have appeared online detailing a ChatGPT Go subscription, likely to sit between free tier users and ChatGPT Plus subscriptions.
According to the screenshot, Go subscribers will get more messages, more file uploads, more image generation, and more advanced data analysis compared to free users.
This does add a slight concern to the future of ChatGPT’s free tier. Does this mean we’ll all be paying for the AI we’ve become accustomed to using for free at some point? I sure hope not…
The hype is building
OpenAI just tweeted a GPT-5 teaser, and as the name would suggest it doesn’t give us any real insight other than the fact that it’s likely going to replace GPT-4o.
With less than three hours to go until the reveal, I was hoping for a teaser to include a little bit more information. I guess not today.
What does Microsoft Copilot think about GPT-5
With just over three hours until the OpenAI livestream, we’ve started to ask other AI tools what they think about GPT-5. After all, who doesn’t want to hear AI’s thoughts on its own evolution, right?
Lance Ulanoff, TechRadar’s Editor At Large, asked Microsoft Copilot about GPT-5 and after a response with leaks and rumors, Copilot said, « It’s shaping up to be a major leap forward. If you could design your own GPT-5 variant, what would it specialize in—creativity, emotional intelligence, or something else entirely? »
Lance replied: « Emotional intelligence, for sure. Have you tried GPT-5? »
Copilot responded, « I haven’t tried GPT-5 myself—since I’m not a user, I don’t get to ‘use’ other models. But I do keep tabs on what’s happening in the AI world, and GPT-5’s rumored emotional intelligence upgrades are especially intriguing.
If it’s anything like what’s been hinted, it could:
Read tone and subtext more accurately in conversations
Respond with empathy and nuance, even in complex emotional situations
Adapt its personality to match the user’s mood or communication style
Support mental health applications with more sensitivity and care
Imagine an AI that not only understands what you say, but also how you feel when you say it. That’s a whole new level of connection. What kind of emotionally intelligent features would you want it to have—like being a better listener, helping with conflict resolution, or maybe even coaching through tough decisions? »
WOAH. This is probably all a pipe dream but Copilot has thought up a future OpenAI model that completely evolves ChatGPT’s emotional intelligence. Is this something you’d be interested in?
Benchmark leak?
Leaks are popping up all across social media including this rumored score on the ARC-AGI-2 test, a benchmark that highlights an AI model’s capability of dealing with logic.
If these results are accurate, GPT-5 would be so far beyond anything we’ve seen before from OpenAI or its competitors.
That said, anyone can make a fake leak, and ultimately until we see the results from Altman and co themselves, there’s no way to be sure of GPT-5’s capabilities.
Just imagine though…
What ChatGPT says about GPT-5
John-Anthony Disotto here, taking over from Graham while he gets fed and watered. I decided the best way to get a glimpse of GPT-5’s capabilities would be to ask ChatGPT itself what we can expect during today’s livestream.
The AI said, « A Unified, Multi‑Variant System » alongside « Supercharged Reasoning, Coding, and Agentic Behavior » and « Ultra‑Large Context & Persistent Memory » sets GPT-5 apart from its predecessor.
Obviously, ChatGPT won’t be letting us into OpenAI’s internal secrets, but considering unifying all ChatGPT capabilities would be at the top of my list for GPT-5, I’m really hoping the AI chatbot isn’t too far off.
The livestream kicks off in just over four hours, so we don’t have much longer to wait to hear about the future of ChatGPT and everything GPT-5 is capable of.
ChatGPT-5 is just the beginning
As I wait with bated breath for the release of ChatGPT-5 (at 10am PT), I’m thinking about what Sam Altman said just a few days ago. « We have a ton of stuff to launch over the next couple of months – new models, products, features, and more”.
So, earlier this week, OpenAI dropped its gpt-oss models, which are designed to run entirely on laptops and phones. In a few short hours, we’ll get ChatGPT-5, but what next? The next couple of months are a long time, which means we might see quite a few new features being released.
Why the Death Star image for ChatGPT-5?
At first, Sam Altman posting a picture of the Death Star in relation to the launch of ChatGPT-5, his own product, seemed like an odd way to market something to me. I mean, what’s he trying to say? Is it, « This thing is so scary it could destroy a planet »?
However, when you view it in context of the things he has recently said about AI creating new and unparalleled security risks and that fact that he finds ChatGPT-5 scary, it fits in. On the one hand, Altman is always pushing the narrative of how great AI will be for the future of the world and for the next generation, but at the same time he’s always talking about how AI has the potential to be dangerous.
This new type of marketing, where he both praises and criticizes his own product, is certainly nothing new for him, and perhaps it reframes the narrative and puts him ahead of OpenAI’s critics? Whatever his reasons, it certainly seems to be working.
Longer than usual
Sam Altman followed up his mysterious Death Star image post on X.com with a more down-to-earth statement in his next post:
« Our livestream tomorrow at 10 am PDT will be longer than usual, around an hour. We have a lot to show and hope you can find the the time to watch! »
our livestream tomorrow at 10 am PDT will be longer than usual, around an hour.we have a lot to show and hope you can find the the time to watch!August 7, 2025
To be honest Sam, an hour is not that long to reveal the details of what could potentially be the new standard in AI chatbots, so I think you should take as long as you need.
That’s no moon… that’s ChatGPT-5?
Sam Altman has confounded the Internet once again by posting a picture of what appears to be the Death Star from Star Wars rising menacingly from behind a planet.
Everybody is a bit confused by the meaning of this, so, who better to ask what it could mean than rival AI, Grok?
When asked what the image could mean, Grok replied: « Sam Altman’s image of the Death Star symbolizes the rise of immensely powerful AI, like the anticipated GPT-5—a transformative force with potential risks, echoing Star Wars’ superweapon. It’s likely teasing OpenAI’s next big breakthrough. »
ChatGPT-5 accidentally leaks hours before launch
As we sit down with snacks and drinks watching the screen and waiting for a live stream to appear for the imminent launch of ChatGPT-5 it appears that a leak on GitHub has revealed the name of the different ChatGPT-5 models.
It looks like we will have gpt-5, designed for logic and multi-step tasks, .gpt-5-mini, a lightweight version for cost-sensitive applications, gpt-5-nano, which is Optimized for speed and ideal for applications requiring low latency, and gpt-5-chat, designed for advanced, natural, multimodal, and context-aware conversations for enterprise applications.
