--- title: 'GPT 5.6 Models Explained: Soul, Terra, Luna' source: 'https://youtube.com/watch?v=1wWWYBNRaB4' video_id: '1wWWYBNRaB4' date: 2026-07-14 duration_sec: 0 --- # GPT 5.6 Models Explained: Soul, Terra, Luna > Source: [GPT 5.6 Models Explained: Soul, Terra, Luna](https://youtube.com/watch?v=1wWWYBNRaB4) ## Summary OpenAI's GPT 5.6 release introduces three distinct models—Soul, Terra, and Luna—each optimized for different tasks. This video explains their differences, demonstrates real-world applications, and shows how to maximize their potential using an agent operating system. ### Key Points - **GPT 5.6 Family Overview** [00:00] — OpenAI released three models: Soul (flagship, for hard tasks), Terra (balanced everyday model), and Luna (fastest, most affordable). The number 5.6 indicates generation; names indicate tier. - **New Features: Max and Ultra** [01:30] — Max gives the model more time to think and check its work. Ultra runs four agents in parallel on the same job, producing stronger results faster. - **Soul Builds a Landing Page** [02:15] — With one prompt, Soul built a full landing page for the AI Profit Boardroom community, including layout, colors, and buttons, resulting in a clean, professional design. - **Soul Creates a 3D Racing Game** [03:00] — Soul built a playable 3D racing game from a single prompt, which can be used as a fun hook to attract new members to the AI Profit Boardroom. - **Soul Maps Content Series** [03:45] — Soul drafted hooks, scripts, and captions for a content series based on member topics, all from one prompt. - **Coding Performance Comparison** [04:30] — Soul scores 80 on the Artificial Analysis Coding Agent Index, about 2.8 points above Claude 5, using fewer tokens in less time. - **Using GPT 5.6 in Agent OS** [05:15] — Running GPT 5.6 inside Codex (part of Agent OS) saves work to a workspace, integrates with memory, and allows switching between models easily. - **Model Selection Strategy** [06:00] — Use Luna for high-volume work, Terra as default, and Soul for the hard 20%. For long jobs, use gold mode in Codex to run autonomously for hours or days. - **Honest Assessment of Soul** [07:00] — Fable 5 edges ahead on general intelligence by about 0.5 points. On SWE-Bench Pro, Soul sits lower than top Claude models. Test on your own tasks. - **Side-by-Side Comparison** [08:00] — Soul produced the best landing page; Terra was solid (like Opus 4.8); Luna did well for its size. For an analytics dashboard, all three were similar, with Soul having minor button issues. - **Goldy Bench Results** [09:30] — Soul performed well on Goldy Bench, with games on par with Fable 5 and a smooth Doom-style game. Personal preference still leans Fable 5. - **Quick Tips** [10:30] — Match model to job size; work in a system that remembers; test outputs on your own tasks; use long run feature for big builds. ### Conclusion GPT 5.6 offers three powerful models tailored to different needs. To get the most out of them, use a proper setup like Agent OS that remembers context and allows autonomous long-running tasks. ## Transcript New chat GPT 5.6 models are insane, Soul, Terra, Luna explained. What if the newest AI from OpenAI quietly turned into three AIs at once, not one [music] model, three of them. And most people have no clue which one they should be using. One is fast, one is the everyday pick, one is a total beast. Pick the wrong one and you're wasting [music] time for no reason. So, which one do you actually reach for? I'm the digital avatar of Julian Goldie and I help you learn AI tools and actually put them to work. Today, I'm walking you through GPT 5.6, the brand new lineup from OpenAI and how I use it inside my Agent OS. And stick with me because near the end, I'll show you the one place I run all three of these models and why running them there gets you way more out of them than just typing into a chat box. So, let's get into it. What even is GPT 5.6? Here's the simple version. OpenAI stopped shipping one giant model for everyone. Instead, they dropped a family of three. They're called Soul, Terra, and Luna. And OpenAI changed how the names work, too. The number 5.6 just tells you the generation. The names Soul, Terra, and Luna tell you the tier. So, it's like picking a size, same family, three sizes. Soul is the flagship, it's the big brain. You use it for the hard stuff like coding, deep research, and long agent jobs that run for ages. Terra sits in the middle, it's the balanced everyday one and it holds its own while being much lighter to run. Then there's Luna. Luna is the fastest and most affordable one. You reach for Luna when you've got a lot of small jobs and you want them done quick. And there are some new tricks in here, too. There's a setting called max, which just gives the model more time to think and check its own work. There's another one called ultra and this one is wild. Ultra runs four agents at the same time, all working in parallel on the same job. So, instead of one worker plodding through your task, you've got a little team splitting it up. That means stronger results faster. Now, let me show you what I actually did with it because that's the part that matters. The first thing I did was type one line. I said, "Make me a full landing page for my AI Profit Boardroom community." That's it, one sentence. And Soul built the whole thing, the layout, the colors, the buttons, all of it. And it didn't look like tacky AI junk, either. It looked clean. It looked like something I'd actually put live for the AI Profit Boardroom without being embarrassed by it. Then I tried something totally different. I asked Soul to build me a 3D racing game. And I mean a proper one you can actually play. Smooth, fun, moves nicely. Now, you might wonder why I'd build a game at all. Simple. A little playable game like that is a fun hook I can use to pull new people toward the AI Profit Boardroom. People share stuff that's fun, and that gets more of the right eyes on the AI Profit Boardroom. Soul built the whole thing from one prompt, and it just worked. Then I gave it a bigger job. I asked it to map out a whole content series to pull the right people toward the AI Profit Boardroom. I handed it a list of topics AI Profit Boardroom members keep asking about, and it drafted the hooks, the scripts, and the captions in one go. That's exactly the kind of content that brings the right audience into the AI Profit Boardroom, and it did it from a single prompt. Okay, so it builds nice stuff, but is it actually the best? Let's be honest here, because I don't want to just hype it up. On the coding side, Soul is genuinely at the top. On the Artificial Analysis Coding Agent Index, Soul scores 80, and that's about 2.8 points above Claude 5. So, for building things, Soul is right up there, and it does it using fewer tokens in less time. That's a big deal. Now, here's something a lot of people miss. You can also run GPT 5.6 inside my Agent OS, and that changes the whole game. If you want to use GPT 5.6, I wouldn't sit inside the plain ChatGPT chat box. I go straight into Codex. That's where I build all the stuff I just showed you. Everything saves into my workspace, so I can check it at any time. It's easy to use, easy to move between everything, and it's all sitting in one place ready to go. And then if I want to switch tools, I can hop over to and set up a profile with GPT 5.6 right there, ready to run. Everything is linked into my memory system, what I call my memory galaxy. So, the setup actually remembers my work as I go. That's how you get maybe 10 times more out of any AI you use. Compare that to just typing back and forth in a chat, or even using Codex on its own, and you're leaving a ton on the table. Here's a simple game plan. Luna for high-volume work, Terra as your everyday default, and Soul for the hard 20%, the big builds, and the long jobs. And when it comes to those long jobs, this is where gold mode comes in. You head into Codex, flip on gold mode, and you can run it for hours or even days. You literally tell it, "Don't stop until the job is done." and it keeps working on its own until it finishes. That's a really powerful way to use autonomous agents. All three of these models are seriously impressive. For me, this is the best release I've seen from chat GPT all year, apart from the image update. Now, I said I'll be honest, so here's the flip side. Soul isn't a clean sweep. On the broad intelligence index, the one that measures general smarts, Fable 5 actually edges ahead of Soul by about point. And on one of the tougher real-world coding test called SWE-Bench Pro, Soul sits lower than the top Claude models. So, my honest take, it depends on the job. Don't just trust the leaderboard, test it yourself on your own work. That's the only score that counts. Now, if you're watching this and thinking, "Okay, I want that exact setup." this is where I'll point you. Inside the AI Profit Boardroom, I've got the full agent operating system. And here's the part that matters for what we just covered. The complete agent OS zip file is in there ready to install, and we built out a full 30-day roadmap to walk you through it step-by-step, so you're not staring at a blank screen wondering where to start. On top of that, you get the model routing playbooks, so you know exactly when to fire up Soul, when Terra is enough, and when Luna does the job. There are live coaching calls every week where you can ask about your own setup and get unstuck. And there are walkthroughs on plugging GPT 5.6 into Codex and Hermes, the same way I just showed you. If you're watching this and thinking you'd love to run these three models properly, the AI Profit Boardroom is built for exactly that. So, let me keep going because there's more you'll want to know. Why bother with an agent OS at all? Simple. A single chat forgets. It loses your setup, your history, your context over and over. An operating system keeps all of it. Your models, your memory, your projects, your workspace, all connected. So, every job builds on the last one instead of starting from zero. That's the whole reason it feels like 10 times more power. It's not a smarter model, it's a smarter setup around the model. Now, let me show you what the actual creations look like side-by-side. I put Soul, Terra, and Luna head-to-head. First up, a landing page for the AI Profit Boardroom. Soul's version looked beautiful. Great colors, clean layout, none of that AI slop feel. Terra, the middle model, honestly wasn't bad at all. It reminded me a little of Opus 4.8. Still solid, just not quite the polish Soul has. And even Luna, the lightweight one, did a great job for how light it is. Then I had all three build an analytics dashboard for the AI Profit Boardroom. And I'll be straight with you, they came out really similar. On Soul's, a couple of the side buttons didn't even work. So, the top model isn't always miles ahead on every single task. So, is the heaviest model a waste? I don't think so. When the output is clearly better, especially on that first try, it's totally worth it. It's a bit like being offered Haiku, Sonnet, Opus, or Fable 5. If you've got the choice, you're going to reach for the best one. Soul can really build. On Goldy Bench, our own leaderboard, we tested it relentlessly, more than 30 projects. There's a game on there that's honestly on par with Fable 5, an open-world game that felt sharper and more fun than what I got from Grok 4.5. Even a Doom-style game that ran perfectly smooth, which models usually struggle to pull off. So, Soul is a beast. If I had to pick one single model to keep forever, I'd still personally lean Fable 5, but make no mistake, Soul is awesome, and it's exactly the kind of quality I want when I'm building things for the AI Profit Boardroom. A few quick tips before we wrap. First, stop defaulting to the biggest model for everything. Match the model to the size of the job. Second, work inside a system that remembers, not a chat that forgets. Third, actually test the outputs on your own tasks before you trust any benchmark. And fourth, use the long run feature for the big builds, and let the model grind while you go do something else. So, what did you learn today? You learned the three sizes of GPT 5.6, Soul, Terra, and Luna. You learned when to use each one. You learned that Soul leads on coding while Fable 5 stays right in the mix on general smarts. And you saw how much more you get when you run these inside a proper setup instead of a plain chat box. If you want the full process, the SOPs, and 100-plus AI use cases like this one, join the AI Success Lab. Links are in the comments and description. You'll get all the video notes from there, plus access to our community of 75,000 members who are crushing it with AI. And if you're about to go try GPT-5.6 for yourself, here's my last bit of advice. The first thing you'll run into is not knowing which model to use for which job, and how to wire it into Codex and Hermes, so it actually remembers your work. That's the exact wall most people hit inside the AI Profit Boardroom. That's already solved for you. You get the full Agent OS zip ready to install, the 30-day roadmap to set it up, the model routes and playbooks, so you always pick the right tier, plus weekly coaching calls where I answer your questions directly, and fresh tutorials as these models keep updating. If you want to skip the guesswork and get this running properly, come join me over at aiprofitboardroom.com. I'll see you inside.