Opus 5.5 vs Sonnet 5.5: Same Price, Very Different Game.

Author: Leon van Zyl · Watch on YouTube

Video authored by the channel shown, not by IANIA; all rights belong to the creator. Transcript and summary generated automatically; the original-language content is preserved.

Original language: English

Original video: Opus 5.5 vs Sonnet 5.5: Same Price, Very Different Game. · Channel: Leon van Zyl · 10:46 · Watch on YouTube

Topics: AI agents, AI evaluation and benchmarks, Software engineering with AI

Summary

The video compares Fable 5.1, Opus 5.5, and Sonnet 5.5 by asking each to build the same browser-based 3D mini golf game, then evaluating the results, completion time, and cost. Fable produced the most polished experience in the review, but took about 1 hour and 40 minutes and cost about $51. Opus and Sonnet took similar times and cost nearly the same, at about $27 and $26 respectively; the presenter judged Opus’s game notably better than Sonnet’s. The presenter concludes that, for this task, Opus is the preferred choice at the same approximate cost and completion time.

Key points

The same open-ended prompt asked each model to create a polished 3D mini golf game with nine courses, animated obstacles, ambient details, and sound.
The models were run with extra-high settings; the presenter notes that Claude does not recommend Max Mode with Sonnet 5.5.
Fable 5.1 took about 1 hour and 40 minutes and cost about $51.
Fable’s game included working ball physics, animated environments, a windmill with collision, and a dinosaur obstacle.
Opus 5.5 took about 1 hour and 26 minutes and cost about $27.
Opus delivered a detailed game with working physics and interactive windmill and dinosaur obstacles, though the presenter noticed the putter followed the ball.
Sonnet 5.5 took about 1 hour and 20 minutes and cost about $26, close to Opus in both time and price.
Sonnet’s game worked, but the presenter noted wonky animations, clipping through the windmill, and a blocked-off hole near the dinosaur.
The presenter preferred Opus over Sonnet, saying Sonnet would need more work to reach Opus’s level.

Chapters

Start	Chapter	Summary
00:00	The comparison and shared prompt	The presenter introduces the three models and sets up a test using the same mini golf game prompt for each.
03:34	Fable 5.1 results	After checking the agents’ progress, the presenter reviews Fable’s cost, completion time, visuals, physics, and animated obstacles.
06:07	Opus 5.5 results	The presenter tests Opus’s game, highlighting its detailed graphics, functioning obstacles, and a minor putter movement issue.
08:11	Sonnet 5.5 results and verdict	The presenter compares Sonnet’s cost, speed, and game quality with Opus, then concludes that Opus is the better choice for this task.

Quotes

benchmarks can say one thing, but the real world use case is where it matters. — 00:00

Fable took about 1 hour and 40 minutes, and this cost about $51. — 04:04

Opus 5.5 took about one hour and 26 minutes and cost about $27. — 06:07

The session also cost about $26, which is only like $1 away from Opus 5.5 — 08:11

for the same price and for the same time, I would rather just use Opus. — 10:15

Concepts and entities

Opus 5.5, Sonnet 5.5, Fable 5.1, Claude Code, Anthropic, Agentic Labs, Jev, TypeSafe AI, minigolfe 3D no navegador

Full transcript (original language)

00:00 I’ve spent hours playing with Sonnet 5.5 today, and it’s a really impressive model. According to the benchmarks, Sonnet 5.5 gets very similar scores to Opus 5.5, especially on agentic coding tasks. And Sonnet 5.5 is about half the cost of Opus 5.5 and about 5 times cheaper than Fable. Now in my experience, benchmarks can say one thing, but the real world use case is where it matters. So in this video, I’m going to compare the outputs from both Sonnet 5.5, Opus 5.5 and

00:31 Fable 5.1 by giving them the exact same prompt. We’ll then compare the quality of the output, the speed and the cost of each of these projects. First, let’s kick things off with Fable. I think this is the perfect model to set the scene. So what I’m going to do is select Fable 5.1, and I’ll run each of these models on extra high. Now this is an important point. Claude themselves do not recommend using Max Mode with Sonnet 5.5. It actually reduces the quality of the output, because I think

01:02 what happens with these models is they tend to run too long, and they kind of start to do more work than what was required, and they even start gaslighting themselves. Then let’s throw in a prompt like “Build a complete, polished 3D adventure mini golf game that runs in the browser.” I’m just giving a little bit of details on the tech stack that should be used, the gameplay. We want 9 courses in total, and for at least two of the courses I want a little windmill with animations, and an animated dinosaur on one of the holes.

01:32 And then I’m asking the agent to design the other seven holes itself to see how these models can actually reason, and how creative they can be. Fill every course with props and ambient animations so it feels alive, like waterfalls, swaying plants, little critters, flickering torches, spinning gears and so on. And for the visual style, we’ll go with cartoonish and a colorful stylized shader look. The agent should also add sound, and that’s pretty much it, and I’m just explaining what done means.

02:02 If you want to try out this prompt yourself, you can download it from the free community. So let’s send this. While Fable is running, I’ll open another folder for Opus. Let’s select Opus 5.5 and on Extra Reasoning. I’ll paste in the exact same prompt and send this off as well. And finally, let’s do Sonnet as well. I’ll paste in the same prompt. Let’s select Sonnet 5.5. And if you don’t see Sonnet, all you have to do is update your Claude Code instance, so whether that’s a CLI tool or desktop app,

02:33 and you should see it immediately. I’ll put this on Extra Reasoning effort, and let’s send this prompt. To be fair, I do want to mention that Sonnet is really good if you give it very precise instructions. An open-ended prompt like this can be tricky for the model to implement, but that is perfectly fine. We’re trying to push the limits of what is possible with this model. Now we’ll just wait for these agents to complete, and then we’ll have a look at the quality of the games, we’ll compare costs, and how long it took for each game to complete. If you would like to take your agentic coding skills to the next

03:03 level, then consider joining my community at Agentic Labs. This will also give you access to several classrooms in my agentic coding masterclass, from the fundamentals of working with coding agents, building design systems, and several projects that you can follow along with. We also have weekly live sessions, like on Tuesdays we’ve got a very casual coffee chat, and every Thursday we’ve got a deep dive session, and this week we’re actually discussing Jev from TypeSafe AI. All of these deep dive sessions are recorded, so you can always go back and watch them again. I’ll leave a link to Agentic Labs in the

03:34 description of this video. I do hope to see you there. It’s been about 30 minutes, so let’s just see how everyone is doing. On the Fable side, it’s still clicking away. I don’t really see anything in the browser yet. For Opus, it’s still working as well, and I don’t see anything in its browser either. And for Sonnet, we actually already have a screenshot, and we do have something running in the browser. So, all three of these agents are making progress. It seems like Sonnet might be ahead, but through the power of editing, you’re going to see

04:04 all three results in three, two, one. All three agents are done, so let’s have a look at their games, and for each of these, we will have a look at the costs as well. Here are the usage stats for Fable. Fable took about 1 hour and 40 minutes, and this cost about $51. And this is what Fable came up with. So far, the menu looks really cool, the background, the graphics, everything does look like these cartoony cel shader looks, so everything looks cool. There are sound effects as well. I’ll just leave them enabled, but I’ll just turn the volume down so

04:35 they’re not too distracting. For the graphics, we’ll just set it to “High”. Then, we can actually view instructions on how to play the game. We can choose a hole, and we can just start playing. So far, the animations look really cool, there’s like a ton of little details in the world. Not sure if you can see it, but the grass is like moving in the wind, we’ve got little butterflies and bunnies. But of course, let’s see if we can play the game. The ball physics look really cool as well, and man, we almost got that into the hole.

05:05 There we go. So this all works, it looks beautiful, but let’s see if it was able to create the holes with the windmill and the dinosaur. So let’s go to our windmill map, and the animations look brilliant. And our windmill is indeed there, let’s see if we can actually putt through it. We do have collision with the windmill as well, so that is definitely all working. Let’s try to putt through this. I’m really terrible at this, so let’s just give it one more shot. And there we go, we’re on the other side.

05:36 So let’s actually just see if the dinosaur works. Let’s go to “Dino Valley”. Oh man, alright, so the dinosaur is indeed there, and it’s probably going to be a bit of an obstacle to get past. Let’s see then. And its tail almost hit the ball. What happens if we crash into its egg? Nothing. Would have been really funny if the egg actually hatched. This is really, really cool. By the way, if you want to play these games, I’ll link to them in the free community along with their source code. But honestly, I think

06:07 Fable did a fantastic job. Next, let’s have a look at Opus. Opus 5.5 took about one hour and 26 minutes and cost about $27. It’s nearly half of what Fable cost. But of course, let’s have a look at the game itself. The menu looks really cool. We do have instructions on how to play the game. We can select that hole. We can select the different graphics options, and we do have sound. Let’s start the first hole. The graphics look really cool. We do have that cel shader cartoonish look to it. And there is a ton of detail still with

06:38 the grass, the little bunnies, the butterflies. This does look really cool. So let’s see if we can actually play the game. So this looks really cool. The ball physics do indeed work. I do notice that the putter actually moves along with the ball, which is not quite correct. But I’m sure Opus can easily fix that with a single prompt. Let’s have a look at the windmill and the dinosaur maps. So let’s go to windmill. The animations do look really cool. And let’s have a look. Hopefully, we can get past it this time, which we do.

07:08 The ball physics work extremely well. And are we good? Man, that was close. Right, so let’s go in. Cool. And of course, let’s have a look at our dinosaur map. Let’s go to Jurassic Jungle. This looks really cool. We do have our dinosaur. And the dinosaur looks super cool. Let’s just see if we can get past this dude. Does the tail have any physical collision? Yes, it does. And the dinosaur just hit the ball into the water. Well, freaking done. So I think this was a really good effort.

07:40 Everything works. The graphics look cool. We’ve got very accurate sort of physics. I’m super impressed. And let’s have a look. What other maps do we have here? So the visual variety is truly impressive. I’m not even sure if this map would be possible, but I’m going to try. Man. So please let me know down in the comments which version of the game you think looks best. And now for the moment of truth. Sonnet 5.5. Let’s see if these claims from Anthropic are true. So Sonnet 5.5 took about

08:11 one hour and 20 minutes. This is very similar to Opus 5.5. The session also cost about $26, which is only like $1 away from Opus 5.5, which is a significantly more intelligent model. So I’m not really seeing much of a performance benefit or even a cost benefit here. This really does come down to the quality of the game after all. So this is the game that Sonnet 5.5 produced. The menu does look cool. The graphics definitely look like they’re shader graphics. And we do have sound as well.

08:41 We’ve got instructions on how to play the game. We can choose the different holes. This is very similar to the other models. Let’s try to play the first round. And the animations do look a little bit wonky. I can also hear music, by the way. I do think I’m just going to disable the music as it is quite distracting. This was the only model to actually add music and not just sound effects. Firstly, let’s see what the physics are like. So when we hit the ball, the putter seems to follow the ball as well, just like– which is the

09:12 same bug that Opus introduced. But hopefully it can fix it quite easily. Can we actually get the ball into the hole? Let’s try again. Oh my word. Let’s try to get it in this time, please. And we’re in. Cool. So everything seems to work. The ball physics work. We’re able to actually get the ball in the hole. So let’s have a look at the windmill. We are clipping through the windmill, so that’s a little bit of a bug. But let’s see if we can actually get through the windmill.

09:43 And if there’s any sort of collision. Let’s try to get through. I do notice that you actually have to hit the ball really hard to move it in this game. So I do think the animations need a little bit of tweaking. But let’s actually go to our dinosaur. And I think that’s the most derpy-looking dinosaur I’ve ever seen. And I noticed the hole is actually blocked off as well. So I’m not sure how we’ll get through it. Let’s actually just whack this ball. And we’re past the dinosaur. Does it still have collisions?

10:15 Yes, it does. And we’re outside of the map. Now, what do you guys think? Which version of the game do you prefer? I must admit, I was expecting Sonnet to be a lot cheaper and a lot faster. But it ended up costing the same as Opus 5.5. And I don’t think the quality is quite the same, to be quite honest. I would much rather prefer Opus 5.5. Although Opus also had mistakes in its game, I think it’s kind of easy to get it to Fable’s level with one or two additional prompts. But I do think that Sonnet needs a lot

10:46 more work to get to the same level as Opus. And for the same price and for the same time, I would rather just use Opus. If you enjoyed this video, hit the like button and subscribe to my channel for more agentic coding tutorials. I’ll see you in the next one. Bye bye.