S2. Ep27 - GPT 5.6, ChatGPT Work & GPT Live: Every New OpenAI Update Explained

Katie (00:26)

Hello and welcome back to another episode. Hi, hello, I'm Katie. And as always, I've got Noel here with me today. Hi Noel, how are you doing?

Noel (00:35)

Yeah, I'm doing absolutely fantastic this week. How are you doing?

Katie (00:39)

Yeah, all good, thank you. Still riding through this very odd UK summer where it's actually hot the whole time.

Noel (00:49)

Yeah, where's the rain? Come on.

Katie (00:52)

I don't know. Probably when the children break up. It's probably coming. So we usually start the podcast off with any updates from the AI and the automation world, but this week there have been so many updates from OpenAI.

Katie (01:20)

We thought we would talk about them and discuss them throughout the whole of this podcast episode. There are a lot of updates. We're obviously only going to be talking about the ones that really affect people on a day-to-day basis, rather than all the little behind the scenes things that we don't need to know about perhaps. But yeah, we're doing a whole podcast episode on the new updates as of July 2026 from OpenAI. Because there's a lot, isn't there, Noel?

Noel (02:05)

So many, it was ridiculous. Again, it was after I'd edited the last episode that they all came through, and I was like, wow, there's a lot. And there's some really good stuff which should be really helpful for most people listening to this. So yeah, exciting times.

Katie (02:11)

Yeah. And we always try and make this podcast suitable for anyone who is listening. So whether you're a beginner or you've got more experience with AI and automations, because we want to make sure that AI and automations are accessible and able to be understood by all people within business, rather than just talking a load of jargon at people, and people having to stop the podcast and go, I don't know what that means, or people just switching off because they don't understand. So yeah, we try and make this suitable for anyone who's listening, whether you've got months of experience or perhaps you're just getting started out and you want to learn more.

Katie (03:12)

So Noel, do you want to start us off with our first new update and we'll take it from there?

Noel (03:19)

So the first part of the updates: I will go through the new models that OpenAI released last week. I thought OpenAI had got around the silly naming convention for AI models — I thought they'd put that to bed six months ago, but clearly not, they're back. So this time they've released GPT 5.6, but there are kind of three variants of that AI model. We've got Sol, we've got Terra and Luna. They all vary on how good they are, what they're good at, and how expensive or cheap they are on the API, things like that.

So Sol is the top tier one. That is the super smart one — I would put that almost in the realms of Fable 5 from Anthropic. In some of the benchmarks, it's actually very, very close in how good it is. Terra is a bit more like Opus 4.8 from Anthropic, so it's the next one down — still very capable. And Luna is something that's really lightweight, really fast, and actually really cheap when it comes to API costs and things like that. So yeah, they're incredible models. I've had a little play around with them, and I would say they're definitely getting a lot better from OpenAI. Over the last six months, I kind of got a bit fed up with their models — I went all in on Anthropic — but I think these ones are starting to make me look again at OpenAI, and maybe start to use them a little bit more in the future, which is kind of interesting.

Katie (04:50)

That's really interesting, because we've talked about this on several episodes, where we've said OpenAI have tried to do so many different things, whereas Anthropic are like, this is what we do and it's just this. And that's why perhaps they were a bit better when it came to their AI models, especially when it came to research. So it's really interesting now that you're swinging back to OpenAI because of these new updates. It's kind of like, just be flexible and adaptable — for a few months you might be using Anthropic, and then in another few months' time you might be using OpenAI, and then who knows, we might swing back to Anthropic again.

Noel (06:02)

Yeah, that could be next week, I don't know. You don't know how quick it moves.

Katie (06:05)

When we were having a chat before we hit record, you actually said to me, Noel, you were thinking of cancelling your Anthropic subscription because of these updates. Can you go into a bit more detail about that? Because you are such an Anthropic fan — like, you and Claude are BFFs.

Noel (06:17)

Yeah, we were. I don't know, it's a little strained. It's a strained relationship at the minute. So with Anthropic, they're dangling that Fable 5 carrot. That's the best AI model you can get your hands on currently. There is another one called Mythos, but that's unobtainable unless you're a big, huge company like Apple — they've got access to it. We're not in that sort of realm yet.

Noel (06:58)

But what they keep doing with Fable 5 — so it got turned off a few weeks ago, then it came back and they said, you've only got it for seven days. And then the day before it was going to get turned off, they said, we're going to give it to you now for another seven days. And that's going to run out in a couple of days' time. And the more I use it, the more I'm thinking, this is actually really slow. It's very clever, it's just really slow.

But I find 5.6 Sol is actually just as clever, and it shows in the benchmarks. It is really, really clever at coding and all sorts of complicated tasks, and it's just so much quicker. So I'm thinking I might as well just go back to OpenAI and use that capability, because on one of the coding benchmarks it's literally one point behind Fable. I'm doing loads of apps and websites at the minute, so why would I not go back to OpenAI and use that? Because they're not threatening to turn it off anytime soon. But I still love Anthropic for the business partner side of things, because it's very honest. It will tell you quite straightforwardly — it's like, no, I don't think you should be doing this.

Katie (08:06)

Okay, yeah, that makes a good point.

Noel (08:26)

With OpenAI models, I just find that they're a bit of a yes man. They'll agree with anything — anything's a good idea, even if it's not. Whereas Anthropic, they've really nailed that down. They're like, no, no, no, this is terrible. Stop. Do not do what you're doing.

Katie (08:41)

Yeah, because that's what was annoying me about OpenAI. It was just agreeing with me, and I didn't want it to agree with me. I wanted it to give me a more balanced view or opinion on things. With all the information that AI can bring to our fingertips, it was kind of like, you can scan the internet within seconds and actually tell me if this is good or not. And it was just bringing back results that it thought I wanted to hear, whereas actually I wanted to be told, no, this isn't great because of X, Y, Z.

Noel (09:25)

Yeah, you need that balance.

Katie (09:39)

Yeah, or if you're thinking of doing something like this, it might be better if you went down this route, or looked at this point specifically.

Noel (09:53)

Yeah, exactly. So I'll still use it for that. But I have not turned any subscriptions off just yet. I've not gone all in.

Katie (09:55)

Okay, maybe we'll check in again with you next week, see if you've still got your Anthropic subscription.

Noel (10:08)

Yeah, I'll be wearing an OpenAI t-shirt or something next week. Who knows?

Katie (10:11)

Who knows? Okay, so that's really, really interesting. And if you're listening to this and you've tried the new OpenAI models, we'd love to hear your thoughts. What do you think about them? You can either send us an email, hello@makeautomations.ai, or come over to our free LinkedIn group, which is AI Automations for Business, and it's free to join.

Katie (10:44)

You don't need to be a client or anything like that. You can just come along and join, and you can have conversations in there with us and tell us what you're building or how you're using AI. What you like, what you don't like, what you think is good, what you don't think is good — all of those things, just conversations.

Noel (11:08)

Yeah, love having those conversations.

Katie (11:10)

What's the next update, Noel?

Noel (11:13)

So the next thing is, alongside these models, they also brought out ChatGPT Work. Now this is kind of where I always thought OpenAI was going to go, because yes, it was great — you had this AI chatbot that you could just have conversations with, and if you wanted it to, it could go off and do things in your inbox, or create things in your CRM, things like that you could connect in.

Now they've really moved into the work mode. They're really targeting business owners and employees who use ChatGPT on a daily basis. So with ChatGPT Work, you can create any kind of document or presentation. You can create visualisations and infographics, all based on your data. And the really cool thing is you can also give it access to certain files and folders. So if you have the ChatGPT app on your PC or laptop, you can say, look, you can only work in this area, this is what you're allowed. Or you could give it everything if you wanted to — that option is there, it's up to you if you want to use it. Personally, I wouldn't, but it is there.

Noel (12:41)

So you could create tasks with ChatGPT Work to say, look, I want to do X, Y, Z this week, can you help me do it? And it's like, okay, well, we need to create a presentation or whatever, and it will go off and do that for you within your brand guides and all that sort of good stuff. So it's a really, really awesome switch. And what I've noticed, now they've implemented ChatGPT Work, is the normal chat that we're all used to and have been using for years is actually a side feature — it's more of an add-on these days. So when you click on chat in the desktop app, it just pops up in a little... almost like a chatbot that you would have on a website, in the bottom right hand corner, and you can use the chat function in there. So they're kind of pushing you more towards using it for work than not for work, which kind of makes sense, because that's where the money's at, isn't it, really? Businesses using it for work.

Katie (13:12)

Interesting. Yeah, but then there are so many people who don't use it for work and just use it for personal things. And there are also things like sports clubs that use it, and other community groups, things like that. Because whenever I see a local event being advertised, it's always an AI poster — always the same sort of layout, the same style of graphic. So yeah, it would be really interesting to know why they've gone for that layout.

Noel (14:17)

I think, at the current moment, from having a look and testing it, that's really only in the desktop app where they make that harsh distinction. I think if you're using it on mobile, and also in the browser, there is a toggle button that says work or chat, and the chat is then a bit more traditional, what we're used to. But if you have it on the desktop app, you're kind of forced to use chat in that little tiny widget in the bottom corner.

Katie (14:48)

Okay, so it is only on the desktop app. And do we know how many people, like how many users, actually use the desktop app rather than just the browser or mobile version?

Noel (15:10)

I don't know. I mean, it's got to be one or two, isn't it? Because didn't they recently say they've now got a billion monthly users? So even one percent of that is quite a big group, isn't it? I would guess quite a lot of businesses would be using it now on the desktop app, because I guess you can then control it a bit more from a security perspective and things like that. There's going to be a lot, I would say, that use it. Especially if anyone's doing the Codex side of things, developing websites or apps or anything like that — they're going to use the desktop app to do that more than the browser.

Katie (15:36)

Yes. So I've just asked the internet how many people use the OpenAI desktop app, and apparently OpenAI does not publicly distinguish between web browser and desktop app users. It just says that they've got 1 billion monthly active users across all platforms — so that's web, desktop and mobile.

Katie (16:21)

And then they're saying that the features within the ChatGPT desktop app draw highly engaged cohorts, such as the Codex and ChatGPT Work environments, and that reaches 7 million active users.

Noel (16:36)

Just a few. One or two — me being one of them. Yeah, it's interesting how it's shifted, I guess. It's interesting to see that in action. But actually, Work is open to anyone on a free account. So I have a ChatGPT free account currently, and I can use Work on there. So if you wanted to give it a go, then you can do, free of charge. I think your usage limits are going to be terrible — you won't be able to do that much — but at least you'll be able to see it in action and give it a go.

Noel (17:05)

But there are some other big features in there. Within Work, we can connect to hundreds, if not a couple of thousand, apps. So you can connect to all of the big CRMs, all of the big email providers, you've got things like Slack or Canva — they're all in there. You've got all of your databases. So when you're using the tasks in Work, the AI will then go off and go, well, we need to close this out in Asana. So it will go off, use the Asana app, and then update everything for you. It's really awesome how that links together and works. And I believe you can have your own personal apps connected in there as well. There is a personal tab, but like I say, I'm on a free account, so I can't create any.

Katie (18:08)

Yeah, I don't have access to that on the free account either. Okay.

Noel (18:14)

No, you can see the option, but you can't do anything with it, which is a bit of a tease. They should really put in a button to upgrade — they've missed that in the UI, it's just showing a blank screen. But I guess from that point, you'd be able to connect into MCP servers where you might have data, or an app that isn't maybe as popular that you use on a day to day basis — you could probably connect that in there as well.

Noel (18:43)

So yeah, really, really awesome. Another really cool feature which I've really enjoyed seeing in action is the browser control. So way back — I say way back, was it last year when they released Atlas? Not even 12 months ago. I think it was late last year, wasn't it? But anyway, they brought out their AI browser called Atlas.

Katie (18:45)

Way back last year! Yes, I remember it well.

Noel (19:13)

And it was a bit slow, a bit clunky... well, it was a cool concept, I'll chalk it down as that. No one uses it. But with ChatGPT Work, when you ask the AI to maybe go off and look something up, it will open up little browser tabs, and you'll be able to read the websites which it's currently going through. So you could be like, oh, that's interesting, I could have a look at that, read it myself and see where it's going. Or it might be looking at a website and you're like, this isn't the direction I wanted it to go in, and then you can change it.

Noel (19:37)

But they've also added in a really awesome little feature where you can open your own tab. So while it's doing that, you could just load up LinkedIn and see what's going on, and that's completely separate — the AI doesn't access that browser tab, just you. So you could be working within ChatGPT while the agent's off doing other stuff with the browser, which I thought was quite cool. You can also make it full screen — you can hit a full screen button so your browser tab becomes bigger. But then there's also the option to hide the left-hand sidebar, so there is true full screen. So it's just like you're using Google or any other browser, which is really interesting.

Katie (20:24)

Yeah, that is really cool.

Katie (20:40)

But does OpenAI track what pages you're opening?

Noel (20:55)

Ooh, that's a good question. I don't know the answer to that. Possibly. They must know where the links are going. They probably don't see the data, but they'll see the URL, I guess, that you've opened.

Katie (21:02)

Yeah, what you're reading and what you're looking at. I'm just thinking of tracking cookies, that sort of thing.

Noel (21:16)

I don't know, I'll have to look into that. Who knows? But I'm sure it's not that. I'm sure they're all above board these days with data.

Katie (21:18)

Yeah, update next week on that one. But yeah, it'd just be really interesting to know.

Noel (21:39)

Yes, I will look that up and give an update next week. Okay, so with ChatGPT Work — yes, it's great. I would say the experience is better on the desktop app, because it can access your files. If you open the browser, you then can't access your local files on your computer, so you can't really use it remotely.

But what they have done within the mobile app is you can remote connect to your laptop or PC that you've got the ChatGPT app installed on, and then you can use your files remotely from the mobile app. So if you want to start tasks while you're out and about having a coffee, you can, which is really quite awesome. So I guess the browser at some point will catch up, but the mobile app will still allow you to do that, which is really quite powerful. Because I always have those moments where you think, oh, I should have started that task before I left. But never fear, I can do it now.

Katie (22:46)

It's quite good as well if you forget, or if you get stuck and you're like, oh, I really wanted to have got that started by now — you get stuck in traffic or you get delayed, anything like that. It's good to have it on mobile.

Noel (22:55)

Yes, absolutely. It's really cool, really awesome. One of the last bits is they've added in a thing called Sites into ChatGPT. I think that's mainly on the desktop app. What these do is allow you to build websites based on your business data. I would say these are kind of more like internal tools, the way I would use them. There is an option to publish and make it public if you want to — so you could create a website for your business and then publish it, which is fine. But if you're creating a dashboard with all your financial details, maybe don't publish that. But you could share it — if you've got a team account, you should be able to share it with team members and things like that.

Noel (24:14)

And I find those visuals are better than if you're using the presentations, like PowerPoint and stuff, which it can do. I find the website versions were a lot better, because it can have a bit more visual control of what's going on and make it look nicer. So it'll be interesting to play around a bit more with Sites in the future.

So like I say, everything right now is on usage limits. They do show you in your settings, in the lower left hand corner if you're using the app or browser, what percentage usage you've got for the current month. So they did do something similar to Anthropic, where they said, we're only going to limit you for the next five hours kind of thing. So it'd be like, well, you've got to wait now until 2pm if you've used all of your usage, and then you can come back at 2pm and you get the next five hours. That's kind of how Anthropic works.

Noel (24:32)

But what OpenAI have done is they've gone, well, that's just silly, we're just going to open it up. So you could use your usage for the entire month — which I guess in a way is also risky, because you could use it all in one day or all in one week, and then you're like, well, now the only option I've got is to pay more to get more usage. So I guess there are pros and cons to it, but I kind of like not being restricted. I could use it for a day and then come back to it next week, whatever, and my usage is still there, which is kind of cool.

Katie (25:36)

Okay, that's good. Yeah.

Noel (25:29)

And as if that wasn't enough — so we've had three new models, we've had ChatGPT Work — they also released GPT Live as well last week.

Katie (25:36)

Okay, what's GPT Live?

Noel (25:55)

So this is an AI voice agent. You know when you chat to ChatGPT on your mobile phone and it comes back, and you have that two-way conversation? It was pretty good, I would say — you could interrupt it. Now they've made it even better. I would say it's more human sounding, more conversational, and it can also do multiple tasks in the background.

Noel (26:23)

So it could be like, hey, I'm planning a trip to go to London at the weekend — could you look up a hotel in this area? Also, I'd quite like to look up what restaurants are in that sort of area, and how can I get there? So that's three or four tasks, and basically the agent will then go off and do those tasks in the background.

Noel (26:51)

And you can still carry on with the conversation and go, well, hold up — which hotels did you find? And it'll be like, oh well, I found these ones for you. And you're like, okay, well, how do I get there? And it will grab that task and go, right, well, this is how you get there — you would use the Circle line to get there on the tube, or whatever. So it keeps all of that information. It's more like you're talking to a human assistant, really, which is really awesome.

Noel (27:22)

But at the minute you can only use it on the browser and, I believe, the mobile app. Most people use it on the mobile app anyway, I think. But they are going to bring it out to the API at some point — I think maybe later this month it was earmarked for. So if you use AI agents, like voice agents answering phones or making calls or whatever, then you could link in this new GPT Live model there. Like I said, they haven't released any API prices yet — it's not out yet — but I'd imagine it's going to be quite pricey. But I also think it will give people a bit more of a human conversation, instead of quite dry AI conversations, which they usually are. So I think this one's going to be a lot more accessible and easier to get on with.

Noel (28:15)

They were actually quite clever in their marketing as well, because they had three ladies — they were probably in the 70s or above kind of age range — and they were just sat there chatting to it and getting it to do stuff. So it was kind of like, anyone can do this. It's not just...

Katie (28:20)

Yeah, they're going for that "anyone can use this". Yeah, okay, I like it.

Noel (28:46)

Yeah, and they were loving it. I'm sure they were paid to love it, but they were loving life in that. But they did show a lot in that advert as well. It showed you those multitask sort of workflows and what it could do. So yeah, I thought it was a pretty cool advert.

Katie (28:51)

Yeah, of course. Excellent actors. Yeah, really cool. That's a lot of updates.

Noel (29:18)

I know, it was a lot to cram in, yeah.

Katie (29:20)

Yeah. And also, you've launched Clyde now, which is your AI agent. How is that going?

Noel (29:27)

Yes, he's doing a great job. I mean, we're not far off 200 agents currently working within the platform, which is really cool. And there's so much AI usage going on — I mean, we're at like 105 million AI tokens spent in the last week. So lots of people are using it, and lots of people are getting lots of really great stuff out of it.

Noel (29:59)

They're trying all kinds of different models. So when we connect into OpenRouter, you could get GPT 5.6 Sol — you could use that, that is in there. Or you could use something really cheap and cheerful. And we've got our own benchmark to show you which of those work best and which don't. So you can use this agent for pennies, pretty much, and still get that sort of extra data working with your apps and all that sort of good stuff.

Katie (30:31)

People are giving good feedback, people are enjoying it.

Noel (30:35)

Yes, lots of good feedback so far. Yeah, really enjoying the app. And we've also released an integration to something called Exa, which is like an AI search application. So you can pop your API key in, and now you can have these really super cheap AI models and give them full web browser access. So they can do tons of research for hours and cost you like a dollar. So why not use that? So yeah, that got integrated this week.

Noel (31:31)

Lots more updates coming soon. But I'm kind of holding off releasing the ability to chat with Clyde on a voice conversation, because I'm waiting for this GPT Live to come out. I want to see what that brings, because I think that would give a lot more interactability and stuff like that with the agents. So hopefully late this month we'll bring it out.

Katie (31:33)

Yeah, okay. And if people want to find out more about Clyde, where can they go? Where can they find you?

Noel (31:44)

So you can head to the website, which is projectclyde.app — all the links are on there. Or we also have a free Skool community as well, which is skool.com/clyde. And that's basically like a support area. I give updates on what's coming up next and all the different features that have been added in, so you get that behind the scenes access, as well as some help if you require it. It's growing quickly, that community, which is awesome to see.

Katie (32:13)

That's great. Well, we really hope you've enjoyed this podcast episode, and yeah, that was a lot of information, that was a lot of updates. So well done if you got to the end. But thank you so much for listening, and we will catch you next time for another episode very soon.