GPT-6 Sol + Luna Just Changed AI Agents

Julian Goldie (0:00)

GPT-6 just got two brand new models, Sol and Luna, and they make your ChatGPT way more powerful. If you keep hitting those annoying usage limits, this is for you. These new models let you get way more done on the exact same plan you already have. They also make fewer mistakes, they give you clearer answers, and they run for way longer without cutting you off. And here's the best part, there's a version you can try completely free, which I'll show you later in this video. Also show you a side-by-side test that surprised me, plus the one model you should actually be using for your daily work. By the end of this video, you'll know exactly which model to pick, so you get the most power without ever hitting a wall. Let's get into it. So we have two new models from GPT-6 today, GPT-6 Sol and GPT-6 Luna to the GPT-6 universe. So these are built on the advances behind Astra 6 If you're wondering, are they more powerful? They're actually not. Come on to why that's in a second. And also right now, the prices are actually 50% lower API prices for Sol and Luna compared with GPT-56.

And if you're wondering, okay, where do you use each one? So Astra is designed for the best frontier level results. Then you got Sol, which is good at complex coding and professional work. And then finally you have fast and efficient everyday work at scale with Luna. So that's how to think about each model. Let's talk about the usage limits. So high usage limits and low costs. So basically what that means is, and I don't know if you've experienced this, but for me personally, when I've been using Astra recently, it runs out of tokens so crazy. I hit the usage limits like three times last week. So if you're using Sol and you're using Luna, then you can get more out of your current subscription from Jack GPT when you're using it inside Codex. Now you might be wondering, OK, how do you get access to this? So when I was actually having a look on Codex, it's not available for me right now. I think this is rolling out for people. So if we check for updates, we're on the latest version. We just don't have the latest model. So I think they'll come later today. However, if we go to Jack GPT, and then I switch to Jack GPT work here, you can see that we've got GPT-6 Luna and Sol available right here. And Sol comes in different sizes here. We've got light and medium, then it will switch to astrolite. So astrolite is more powerful than Sol. And then down here we have GPT-6 Luna, which is less powerful than Sol. So if you want the lightest setting for non-frontier work, just something that's super fast, then you would use Luna instead.

If you have a look inside Jack GPT normal here, there's no real dials for that. It's more focused inside the work section here. So this is rolling out today in Jack GPT work and Codex for Plus Pro Business Enterprise and Edu users. Both are available already in the API. Free and Go users. If you're a free or a Go user with Jack GPT, you can try GPT-6 Luna in the desktop app they're saying. Again, it's not available for me just yet, but I think that will come later.

What's pretty interesting as well is that this is just dropped on the same day the Opus 5 has come out as well. So Anthropic have just launched Opus 5, which is performing at the level of Fable 5.1, but it costs 40% less, which is pretty good as well. So the competition is heating up here. And then you can see the full breakdown of the models on their website. So the two new ones are GPT-6 Sol and Luna. And you might be wondering, OK, what are the highlights of this? What are the main points here? So number one, they're half the price. So these are 50% less than the older versions. They're better at real work. They're better at jobs, for example, at handling business tasks across different apps, writing code, using the computer, et cetera, computer use. They also make fewer mistakes. So Sol gets facts wrong about half as often as the GPT-5 or 6 version. So GPT-6 Sol gets facts wrong about half as often, which is better, less hallucinations. And they talk more clearly. So the answers are designed to be a bit shorter, use less confusing jargon, and are more honest about what the AI actually checked. They're cheaper for long conversations. This is big as well because if you're using GPT-6, for example, right now, quite often, as soon as the context window gets relatively large, the tokens just run out so quickly. Also, they're designed to be more trustworthy. So basically, lie less.