Opus 5 The Greatest Model In The World (Full Test)

0:00Anthropic has never released a model on a Friday, and I was totally blown away when we saw Opus 5 come up in my feed on Twitter, and it absolutely blows Fable 5 out of the water. Supposedly Fable 5 intelligence for half the cost. You can see real quick here agentic terminal coding. It gets a 10% boost on Frontier bench 43.3% knowledge work taking ahead, too. We got agentic search up 3%. We also have computer use up about 4%. Agentic coding itself is about on par with Fable 5 and as well as agentic coding on Frontier code, which does take the cake slightly at 53.4%. So, I'm going to be breaking

0:45down Opus 5 and how it compares to Fable 5 in this video running a few tests, and I'll put the links to those test results down in the description below, but I had never expected this to come out today. So, let's jump right into testing. And if we're getting a Fable 5 intelligence for half the cost, which is what you see here Claude Opus 5 just released $5 and $25 out per million tokens. And if we're seeing what we're seeing in those benchmark results, why would I go back to using Fable 5 again? Like seriously, if this is not benchmarked by Anthropic and it actually performs saying how it performs in agentic tool calling, refactoring code bases, UI design, and it hallucinates less, why am I going to

1:31go back to using Fable 5 again? Because that's only available up to 50% on my weekly usage, then I have to pay for credits. This comes with my subscription, which is why we're going to be testing it out in this video to actually see if it's something that I can use as a Fable 5 replacement. I'm excited. Just to give you a quick rundown on how we're actually testing these two models. So, Opus 5 is going to be set to high effort level. Fable 5 is going to be set to high effort level. So, this is all going to be done inside of Becken, our app that we built. And we just run these benchmarks just to see if it's something that's actually going to be a replacement for Fable 5, and it's going to be maybe a new daily driver if it can perform well. First test is just a simple create a game,

2:16a fantasy MMO, you know, similar to like World of Warcraft. See how it does with game design and see if it follows instructions well. The next one is going to be the second test is going to be producing a remotion video. So, remotion is a skill that you can download and have any of these models, Opus, Fable, Kimmy, whatever, create a demo product video programmatically. So, this has been a great test. I have a couple on my landing page right now, and this is a great way to see how it uses this tool to create an exact copy of what your product is about and see how well it does. The next one is recreating a front end design. So, giving it one of my landing pages from either Easy Flip or Magic Hand or another app that I built,

3:02and just seeing how well does with front end design because that has been the most difficult part when it comes to all of these models coming out is the front end design is always lacking. So, I think those three tests are going to be really interesting to see how well Opus 5 performs over Fable 5. So, I have the first prompt to being sent off, which is building the fantasy game. It's going to use 3 JS, and it's going to actually pull up two terminals, one with Fable 5 and one with Opus 5, and they're going to basically battle it out together, and then we can take a look at the results. So, we are clear to go. Hey Becken, can you tell Aspen that we are ready to start the first test? >> Sent to Aspen. >> Okay. So, once that is cooking, we can check back with the results and just see

3:47how long it takes, how much tokens it costs, and the quality of the output. And I just want to show you this here. I thought it was super interesting. If you look here and just pay attention, Fable 5 Max versus Opus 5 Max, $8.23 for Opus 5 Max, and then you're at $17.32 for Fable 5 Max. This is very interesting and a sign that these models are getting better and they don't need to cost as much as Fable 5 to run. We saw that with Kimmy K3, we saw that with GLM 5.2, even if you're going through the API, you could still get Fable 5 intelligence for less than $17.32 per task and 103,000 tokens per task.

4:35So, this is going to be a very interesting test. I don't know if Anthropic intentionally wanted this to happen where they just kind of cannibalize Fable 5 because now I can see who is going to use Fable 5 if Opus 5 performs this well at literally 40 to 50% of the cost. So, I think that this is an absolute home run if we are seeing what we're actually seeing here. Like I said, the price didn't change, $5 in, $25 out for Opus 5. This is going to be a very interesting test. So, we are good to go. Let's launch this thing. So, if you're wondering what app I'm using, I have built a Becca and out, which is a voice-controlled agentic development environment that communicates into each

5:20pane. So, right here you can see Aspen is controlling Opus 5 and Fable 5 via the built-in MCP. It has built-in vision and voice, so I could just say, "Hey Becca, can you check on what Aspen was doing? I forgot. Give me a quick summary." >> Sure thing. Let me check what Aspen was up to and give you a quick rundown. Aspen's basically got the test setup and rollout logic ready. It's coordinating the benchmarks and finishing checks, making sure both runs are on the same configuration, then waiting for both to be verified. After that, it'll ping when they're all playable and ready to review. >> Yeah, and why I built this was because I was so tired of working in so many different terminals on different

6:06projects, and then having to read through the summary over and over again, forgetting what I was working on. No more, long gone are those days. I'm essentially just building a second brain for all the projects that I'm working on. Back end, idea browser, marketing, this space, school discord, shipping school, marketing easy flip. So, I'm building out these different workspaces, so I'm able to keep my second brain more organized when I'm working on these projects. So, whenever a new model drops, we go through the back end bench workspace. This has rules and recipes that are built on the open knowledge format that Google released a couple weeks ago on how to structure files effectively, so agents can read through large amounts of data to follow

6:51instructions, and that don't really eat up tokens as much as they would if you didn't have a folder framework just like this. So, like just a quick recap on what the Google open knowledge format is and how Becca was built with this is it is a knowledge as a directory of markdown files with a YAML front matter. So, it is essentially every workspace has its own LLM wiki. This paper was written by based off of what Andrej Karpathy had published, and then Google comes out on June 12th, I would say yeah, over over 4 weeks ago, and built this framework that I basically implemented inside of Becca into be able to have their own LLM wikis per

7:38workspace. So, this was a big hit, you know, late last year and I wanted to be able to have my agents start working on that second brain of whatever I'm building and being able to synthesize information much more effectively. And this is kind of how the folders are structured. An OKF bundle is a directory of markdown files representing concepts. So, anything you want to capture including tables, datasets, metrics, playworks, runbooks, APIs, each concept is one file. Then the file path is the concept's identity. So, not without getting too much in the weeds, I just wanted to show you, you know, how I'm trying to actually build tools that I use every single day and have the agents

8:25start to remember the context that I'm using and saying to these agents to build upon. So, this has been just a long time coming for the last 6 weeks since Fable 5 first dropped, I've been working on this. I got it linked down below. You have multiple agents, you have a workbench here where you can manage tasks. You could track your usage even if you wanted to. So, all your usage is tracked, bring in your Grok API key, your subscription, bring in Codex, bring in Cursor, bring in your agents. This is a layer that sits on top of your agents with built-in memory, voice, and vision. So, you also have your own memory modules here that you could track. Also, if you want to spin up a YouTube player and listen to some music,

9:10you can as well. But, I was building this to be able to just keep track of the things that I'm trying to manage. We do have built-in chat where you could share your screen just like Zoom. You know, you want to get into Zoom, you're talking to someone, you're trying to problem-solve, you can now invite a buddy to your workspace and you can problem-solve together. They can even have the option to join your workspace. They bring their own subscription in and you guys can work on your own projects together, which is pretty cool. So, that's what I've been working on and because people have been wondering, "Where did you Where are you working in?" That's what's running these benchmarks, and that's kind of how we've been building over these last few days. So, it's still working on the test. It's been going for 7 minutes. This one's

9:55going for 7 minutes. They're both working on building the game, and we'll report back the results when they're all done. Wow, guys, this is pretty crazy. Fable 5 actually finished before Opus. Pretty interesting. Let's take a look at what happened here. Fable 5 is done, 8 minutes 41 seconds. All done. One thing that needs my eyes, the Fable 5 pane has text sitting in its input line. Open in the browser. Okay, so we are going to open it in the browser and just see what we get here. I have no idea. I have not seen this. I have no idea. So, let's see. I want to show you where we are at with game development now. Opus 5 just blew me away. This is the game and just created it about 25 minutes. Ember Hollow, Chronicles of the Veil, The Old

10:41Road to the Grove has gone quiet. I love it. So, it's really impressive on what it made compared to Fable 5. I mean, the shaders look better. I mean, this is very impressive, guys. Like, you got a map, you got inventory, you got a quest log, speak with the elder Maravain in Ember Hollow. This is impressive, guys. Look at this. I mean, the graphics are substantially better than Fable 5. Yes, it took longer, but the mechanics actually work. Ah, I'm dying. I I am pretty impressed, guys. Game development is never going to be the same. Now is the time that if you want to build a game, building with AI just got that

11:27much easier. Let me show you the output from Fable 5 on how I think, you know, it did finish faster. I just think that it doesn't look as good. Here is Fable 5's output, which took 8 minutes. So, it's still really good. It's just not as good. Shaders are a little lot of jagged edges. It's very plain. So, if you put the two together side by side, I just think that visually Opus 5 looks better. Not not saying this is terrible. Remember, but they're on high effort level. So, also the controls are backwards on this game. But, let me show you the side by side. Here is the full side by side. You can see 25 minutes Opus 5, and then you could see 8 minutes for Fable 5. I mean, if you're going to

12:13get granular here, I think that Opus 5 wins. I mean, hands down. I just think it's just looks better visually, more appealing, better shading, better graphics overall, more details, more assets. You have a quest log that you can actually talk to characters. Comment down below, what do you think? Let's move on to the next test, which is actually going to be the remotion video. So, let's go ahead and start on test number two. Can you tell Aspen to start test number two on a remotion video? I would like the remotion video to be a recreation of the Hey Beacon.ai app. I want them to create a amazing remotion video about how Beacon works and the features. Let them cook.

13:00>> Got it. I'll send that to Aspen as a work request and get things moving. Sent. I've relayed the request to Aspen. Start test number two on the remotion video. Recreate the Hey Beacon.ai app and make it a fantastic explanation of how Beacon works and its features. >> There we go. We'll check back. And right here we have Opus and Fable creating the remotion video. So, they're going at it. Few minutes in. They're set to create a 30-second 1920 by 1080 Becca promo UI rebuilt as an animated React components. So, this is going to be a great one as well as testing the UI design capabilities of Opus 5 to see if I would consider it, you know, creating websites for me or landing pages. This is with no

13:45skills installed and Fable 5 is basically orchestrating these two agents as they go through each test. While we fired up some other tests in the background, our remotion video and our complete UI redesign of our landing page, those are running in the background. I thought I'd pull up the artificial analysis intelligence index. You can see here they just dropped Opus 5 on here. Max just pulling ahead of Fable 5. You could see here 61 to 60. The artificial intelligence index incorporates nine E-vals, terminal bench, side code, humanities last exam, couple other ones in there. And this just proves my point that if we're able

14:31to get Fable 5 level intelligence and output for nearly half the cost, I don't really see any use case for Fable 5 other than maybe checking Opus 5's work. So, if you have some type of, you know, agent that orchestrates the whole scene just like we're doing in Becca bench, Fable 5 is running the benchmarks, but it's also checking the work of Opus 5 and Fable 5 as it continues to pump out these results. So, if you were to use it in in your workflow, say Opus 5 does 90% of the work and then the 10% work can be left for Fable 5 on extra high or max to just double-check the output, the code, the refactoring of whatever code base you're working on before you push it to production. That's where I could see it kind of being used, but I don't see, I

15:17mean, is Fable 5 dead, guys? I did notice this last week the hype had been dying down quite a bit. Kimmy K3 came out that blew everyone out of the water. JLM 5.2 came out weeks before that. The whole GPT 5.6 soul on extra high performing really well even breaking out of their own lab and hacking into hugging face. I mean this is a wild time to be alive, but this is one of the strongest model drops from Anthropic regarding the Opus model because the Opus model 4.8 4.6, 4.5 those were all very underwhelming lackluster releases. And now with Opus 5 this really feels like an actual Opus 4 moment. You remember when we got Opus 4.1? We were

16:02like, "Wow, it's literally one shotting websites. That's amazing." That's why we're running it through these tests to see if this is going to be a daily driver for most people and it's already heading in that direction. It really is. Like why am I going to go Why am I going to go back to use Fable 5? All right, so the remotion videos are done. Let's pull up the comparison and just see how they did. Okay, the results of the remotion video just finished and I have to say it's hard to really pick a winner here. I think they both did a really good job. You can see Opus 5 took twice as long and I think Fable 5 did if not better of a job in half the time. So yes, it costs you know 50 cents more, but if it's able to do pretty much the same thing in half

16:48the time, I do like that. And remember Opus slogan is called Slopus. It has been slow since the dawn of time. It doesn't matter if you're on medium, high, extra high, or max reasoning. It just seems to be slower. And Fable 5 on high has been nice and snappy. Never have I experienced its slowness like I have with using Opus models. So you can see here it did a really good job. It's really hard to pick a winner. I mean there's little details here where Opus 5 did a little bit of a job with like the motion and how it kind of goes through the actual remotion video itself. But it did a really good job compared to the original and if you're not using remotion, you need to be. If you have any kind of startup, download the

17:33remotion skill, have your agents use that because you're able to create beautiful 9 by 16, 16 by 9 product demo videos. So the next thing was the UI design on the website. for Opus 5 to finish up here, but Fable 5 has been finished for a while. So we can pull up Fable 5 and just check it out and see how well it did. And that proves my point. You can see Fable 5 finished this website recreation in 4 minutes. Opus is still going and it's taking 26 minutes. Like what in the heck is it doing? It better be a good website recreation cuz dang, that's a long time. All I'm telling the agent to do is just recreate my heybecken.ai website and you

18:18know, we have some remotion videos here. Um a lot of parallax effects. So this is the original website, right? And here is what Fable 5 produced. Uh single HTML file and it did okay. Um there is no parallax effects and it didn't even like complete the website. So I thought that was really weird. Um it only got to the content creator agent and you know, I think there's like too many vibe coded looks to it. Vibe coded meaning like there's too many H1 above the H2, too many pill icons. I just think that those are dead giveaways. But again, it didn't even finish it. So it finished in 4 minutes, but it only got half a website.

19:04So Opus, whatever it is doing, it better be good because we've been waiting long enough. All right, we finally got the output from Opus 5 and here's what it did. Obviously, there's no parallax effect, but it did create some new re-motion videos. Did it actually finish the website? It did, but these tile cards are not relevant. Still looks a little vibe coded with the gradients, but nonetheless, it finished the website versus Fable 5 gave us a half version of the website, whatever that was. So, all of these benchmark results are going to be posted on backandbench.com. And what I would suggest that you guys do, go on to backandbench.com. I got a link down below. Cast your vote. Vote on

19:50whatever you think is the best output. We're collecting votes from actual humans to see what is giving us the best output. Also, the games are going to be posted in the game section, so you're going to be able to play these games as we continue to test these models and upload the results onto backandbench.com. So, this is just the first iteration of the site. We're tracking all of these new models coming out. We're running them through some tests to see if they're actually going to get the vote of confidence to be able to run these models in production. So, I think the results are unanimous. Opus 5 gives you Fable 5 level intelligence for half the cost. It's going to be my daily driver. It did a great job. Yes, it took longer, but I'm not exactly in a hurry. Like,

20:36what am I in a hurry for? I can have multiple agents working in multiple terminals on multiple workspaces in a loop if I want, and I'm not in a hurry. So, if you're looking for speed, maybe stick with Fable 5 on medium to high reasoning level, and then use Opus for some of those longer running tasks that don't require a time constraint. So, very interesting. If you want to know more about where we're at, what we're doing in the community, you could check out the links down below. We have a free Discord, as well as a free school community. You could check it out, join along, show us your test results, how they were, what you thought. What do you think? I mean, are you still using Kimik 3? Are you going to stick with stable 5? I think OS 5 is a great solution to building effectively, get good quality

21:23output, and not break the bank doing so. So, comments down below, what are your thoughts? I want to know. We'll see you in the next video. Have a blessed day.