Kilo Code DESTROYS Fable 5?

Key Takeaways

So, how do you get Fable 5 level intelligence without Fable 5 itself? Obviously, Fable 5 has been taken down. Nobody can use it. But, what can you do in the meantime whilst you wait for it to come back up? Well, Killo code actually created a recent case study like you see here and they've actually implemented the model and I'll show you what we found today where basically if you have a combination [snorts] of different models planning, then you'll get better outputs and you can get better intelligence side by side. So, let's test it out and see. And the thing that I would say with all this, you know, we saw this with Fusion as well that came out from Open Router recently is like you can try it and just see what you think, you know, I wouldn't pay too much attention to benchmarks. I would just test out yourself. I'll show you what we tested so far. And so, the context here is like Fable 5 was, you know, the most powerful AI ever made public and then it was shut down worldwide 3 days after launch. Now, Killo, according to their research, tried to prove something which is if you hand a cheaper model a good plan, it can build the exact same thing that Fable 5 would, identical, but it requires less tokens or requires less powerful models to actually get the most out of it. Again, this isn't my theory. This is the Killo team that ran the test and published it. So, they had two frontier models plan the same service. Then they made both build the winning plan from identical starts. And this is the key point is all in the planning according to this. So, Fable 5, the most powerful model ever made public, 80% on SWE print bench pro, and a cheap model produced services that were identical down to which individual uses 35% roll out enabled, both passing 15 checks. All right, the only difference here was the amount of tokens and the cost. So, according to the Killo blog, planning with Claude Fable 5 and implementing with GPT produces the same service for 59% less than using Claude Fable for both phases. So, you can get a great plan and then plug it into your system. According to this. So, you get like the identical output but 62% less to build. Across a full plan and build pipeline, that's 59% cheaper, which is a 2.4x gap that scales to roughly, you know, as you can imagine, quite a lot. So, the plan quality, where does the intelligence actually live? And you can see the details here. So, they've actually created this blog on you don't have to use Fable and Mythos to work on the frontier. So, basically, what you can have is one plan in and then you get identical builds out, right? And this is what Kilo are trying to prove. So, they gave the same plan to two different models. Both passed all 15 acceptance checks and then the two services were identical, okay? So, you got the planner, which is like that's where you use the frontier model. The plan decides all and then the agents actually build. So, you can use cheaper agents for building to get identical outputs. Now, why does this work? Well, basically, the model stops mattering as much when you have a more pinned plan. So, a vague plan would leave forks open. So, each model would guess, two models build two different apps. A pinned plan decides every fork, and so there's nothing left to guess with a pinned plan. So, any model can land on the exact same build, right? Vague plan, any model, the output differs. Pinned plan, which decides every fork, any model, same app, which is pretty crazy. Now, it won't be like any model, but you get the point. Like, you could use something less powerful, something cheaper, and then you can build it out better. Now, there's three different moves here. So, the smart model would write the plan, you switch the model in a click, and then the cheap model would build it. And then you can have review agents like check it for bugs and security and that sort of thing. So, you got the plan, the build, review agents that check for uh security or logic and that sort of thing, and then it is shipped. And so, if you look at this, you got the old way and the new way, where, for example, if you plan and build with the best models, that would use up more tokens. If you use the new way, you uh plan, and then you can build with a cheaper model, and that would uh be more efficient. 59% less on some of the tests. And also, if one model gets taken down, like Fable 5, it's okay, you can root the plan that you saved to model B. And so, the plan survives, you just swap the builder, and then you don't have to worry about using one model again. Now, also, the other thing about this is that a good plan is a reusable asset, right? So, you could write it once, and then fan it out to as many like cheaper models as you wanted to, building variations. So, you could have one plan that's reusable, and then build it with multiple different models, and just see which one creates the best. Now, you can do this inside Kelo. Here's an example of how that would work. So, you could say, "Okay, Opus 4.8, write a very detailed plan." And then the builders build it with a cheaper model, for example, like Chimera 1.7. And you might say, "Well, won't like the the worst model mess up the build?" If the plan has already made every decision, guessing where builds go wrong, and a complete plan leaving nothing to guess, will both models build the exact same plan, and will pass all 15 checks, according to this research. So, for example, for actually building, you could use a free model like Nemotron 3 Ultra, uh or a cheaper model like Minimax M3 to actually build it out, and then use a frontier model for actually creating the plan. So, the old way would be like, for example, you point every job at the most powerful model, then you're using the frontier token costs associated with that. And also, as you scale up, we would use more and more tokens. Whereas, the new way is like the genius AI model that you want to use plans it, and then cheaper models build it with the identical output. If a model gets removed, like Fable 5, you swap the builder and keep moving. You use less tokens overall and also the plan is a reusable asset. So, you can build from it from forever. And also as you scale, you'd have less tokens used. Now, some people say, "Well, the best model should do everything." But, the best model is great for judgment. It doesn't necessarily need to do everything for you, especially if you've got like a a fail-safe plan that can go off and build stuff. Other people say, "Well, I need to to wait until one model clearly wins." But, Fable 5 was the best and then it got removed. Other people say, "Well, this sounds technical." So, you can do it with like two prompts inside Kilo, but you could also build like for example a skill around that inside Claude. That would be quite interesting, too. So, you could say to Claude, "Okay, use this model for actually creating the build and then use your most powerful model for creating the plan." So, just to recap, you learned how to potentially, and again, just test it yourself, but potentially get a Frontier model to plan and then a cheaper model to build or a free model to build. So, you keep the same quality, but you use less tokens. You can also run it in one tool with Kilo, but you can basically use the most powerful model for the plan and then build it for free with a free API or free model. And as an example of that, like you might be wondering, "Okay, what free models are there?" So, for example, N2, Nemotron 3 Ultra, you got local models, you have free APIs like Our Alpha as well. Um you could just test yourself, see which one works the best for you. So, if you want the whole system that we use for building stuff, you can get that inside the AR Pro 4 and we have a full agent operating system like you can see here for implementing all of our AI agents working together. The great thing about that is for example, if we're using Hermes, we could use a really powerful model for planning and then that can delegate to sub agents. The same with Claude, the same with for example Fusion is a great example of this. So, this is another API that claims to get Fable level level intelligence by having multiple agents basically come up with the plan or the outputs, then a judge actually fuses it together and then you get one answer together. So, that's another alternative as well, which is fusion where you use multiple models. So, you could have like Opus and Gemini and Grok working together. And then that creates on benchmarks according to them again, just test it yourself, but on benchmarks it would potentially hit, you know, the the same sort of benchmarks that you get with a a frontier level model. So, thanks so much for watching. If you want to get all of my best trainings on this sort of stuff, if you want my agent operating system, you can get that inside the AI Profit Bootcamp. You can post questions inside the community. I personally answer them. So does everyone else inside the community, inside the classroom. You can get access to all my best trainings. So, we have a complete beginner to expert course here. And we also have new daily updates like you can see if you like the advanced stuff. We actually got a whole section on fusion and how to use it and get the best outputs from it to to achieve potentially fable five level intelligence. And then also we have the agent operating system that we update daily and you can get this information inside here, too. We also have a calendar here where you can jump on weekly coaching calls, ask questions, share your screen, meet cool people, building similar stuff. Inside the map, you can meet people in your local area who are using AI agents like you. And this is all available inside the AI Profit Bootcamp. Link in the comments description or go to the AI Profit Bootcamp.com. Thanks for watching.