26:07Jay E | RoboNuggets · Review
I Made Fable 5 and Kimi K3 Build the Same App
Same one-shot prompts, two models, three real builds — a website, an app clone, and a 3D game — scored on intelligence, cost, and time, not just vibes.
July 23rdThe argument in one line.
Topping a public coding-arena leaderboard does not guarantee real-world reliability: a same-prompt, parallel test of Kimi K3 against Claude Fable 5 across four build tasks found Kimi ahead on only one, with repeated errors and a fast-draining usage budget along the way.
Read if. Skip if.
READ IF YOU ARE…
- You're deciding whether to add Kimi K3, or another new leaderboard-topping model, to a coding stack that already includes Claude.
- You use AI coding agents (Claude Code, Kimi Code, or similar) for real production work and want evidence beyond a benchmark score.
- You're curious how a Chinese frontier model's actual coding output compares to Claude Fable 5 on identical prompts.
SKIP IF…
- You want a plain spec sheet or feature list — this is a hands-on results comparison, not documentation.
- You don't use AI coding CLIs at all — the comparisons are agent-workflow specific (thinking traces, YOLO mode, usage limits).
The full version, fast.
A new Chinese model, Kimi K3, launched at #1 on a public coding-arena leaderboard, so the reviewer paid $19 to test it head-to-head against Claude Fable 5 on four identical tasks run in parallel: a voxel sandbox game, a motion-graphics launch video, a synth pad with drum sequencing, and a researched brand website. Kimi K3 won only the voxel game, with denser graphics and a working day/night cycle; Fable 5 won the other three, shipping a synced-audio video, a fuller-featured synth, and a more on-brand, accurately researched website. Kimi K3 also burned through its plan's five-hour usage limit within minutes and returned repeated 404/permission errors after a top-up. The conclusion: a leaderboard #1 doesn't predict which tool is actually worth paying for in daily use.
Free for members
Chat with this breakdown — free.
Sign in and you get 23 free chat messages on us — ask for the hook, quote a framework, find the exact transcript moment, generate a markdown action plan. Bring your own key when you want unlimited.
Create a free account →Where the time goes.
00:00 – 00:24
01 · Kimi K3 model overview & specs
Cold open claiming a new Chinese model beat Claude; parameter count and context window stated.
00:24 – 00:33
02 · Kimi K3 vs Claude Fable 5 coding benchmarks
Leaderboard rank stated as the reason for the test.
00:33 – 00:54
03 · Moonshot AI pricing & signing up
Host signs up for the $19/month entry plan.
00:54 – 01:12
04 · Kimi Code vs Claude Code CLI interface
Host launches Kimi Code with --yolo mode alongside Claude Code/Fable 5.
01:12 – 01:48
05 · Coding challenge 1: playable voxel sandbox game
Identical prompt fired at both tools in parallel; Fable 5 shows a long unlabeled thinking state.
01:48 – 02:46
06 · Coding challenge 2: AI motion graphics video (coded, not templated)
Prompt asks for a fully coded launch video with animated text easing and a soundtrack cue sheet.
02:46 – 03:00
07 · Coding challenge 3: synth pad & drum sequencer
Third parallel prompt: a neon-styled synth pad with drum sequencing.
03:00 – 03:34
08 · Coding challenge 4: AI web design & live search test
Fourth prompt withholds brand info and requires the model to search the web for facts about the host.
03:34 – 04:19
09 · Claude Fable 5 voxel sandbox game demo
Fable 5's VoxelCraft build runs in-browser with block-breaking and a day/night cycle.
04:19 – 05:36
10 · Moonshot AI API usage limits & error issues
The $19 plan's five-hour usage limit is exhausted within minutes; top-up leads to 404/permission errors.
05:36 – 06:58
11 · Kimi K3 voxel sandbox game gameplay review
Kimi K3's VoxelCraft build shows denser graphics and a working sunset-to-night transition.
06:58 – 08:26
12 · Claude Fable 5 Neon synth pad UI demo
Fable 5's synth pad ships a fuller control set (attack/release, randomize, multiple banks).
08:26 – 09:35
13 · Kimi K3 synthesizer UI comparison
Kimi K3's synth replicates the same interaction pattern with fewer exposed controls.
09:35 – 10:27
14 · Claude Fable 5 web design results
Fable 5's researched CreatorMagic website with accurate community/YouTube links.
10:27 – 12:03
15 · Kimi K3 web design & search accuracy review
Kimi K3's version is factually accurate but visually busier and less on-brand.
12:03 – 13:17
16 · Claude Fable 5 AI-generated video results
Fable 5's rendered launch video plays back with synced audio, judged on-brief.
13:17 – 14:05
17 · Kimi K3 AI-generated video comparison
Kimi K3's rendered video has no audio and simpler, less polished animation.
14:05 – 15:12
18 · Final verdict: is Kimi K3 better than Claude Fable 5?
Host tallies 3-of-4 wins for Fable 5 and states he's staying with Anthropic despite Kimi's leaderboard rank.
Lines worth screenshotting.
- Kimi K3 debuted at #1 on a public coding-arena leaderboard for front-end code, ranking above Claude and GPT-class models, with roughly 2.8 trillion parameters and a 1 million-token context window.
- A $19/month Moonshot AI plan's entire five-hour usage allowance was consumed within minutes of running four parallel coding prompts, forcing an immediate mid-test top-up.
- Kimi K3's coding agent streamed visible reasoning almost immediately, while Claude Fable 5 sat in an unlabeled 'thinking' state for close to three minutes before producing any output.
- In a head-to-head voxel sandbox game build, Kimi K3's output was judged more graphically complex than Fable 5's, including a working sunset-to-night lighting cycle — the one task where Kimi came out ahead.
- Across four identical prompts run in parallel, Fable 5 won three of four builds outright: the researched website, the synth pad with drum sequencer, and the motion-graphics launch video.
- Fable 5's motion-graphics video output shipped with a synced soundtrack; Kimi K3's equivalent output had no audio track at all.
- Fable 5's AI-researched website correctly pulled the reviewer's real community platform, YouTube channel, and a separate audio-production business from a single unguided prompt with zero brand information supplied.
- Kimi K3 repeatedly returned 404 'model not found' and permission-denied errors mid-test after the top-up, which the reviewer attributed to server overload from launch-day demand rather than an account-specific issue.
- Despite Kimi K3 topping the public arena leaderboard on launch day, the reviewer's actual money stayed with Anthropic and Fable 5 after the hands-on test, illustrating a gap between leaderboard rank and practical coding reliability.
- The test methodology ran identical prompts on both tools in parallel terminal windows specifically so response speed and first-draft quality could be compared directly rather than sequentially.
What a #1 leaderboard rank didn't tell you about Kimi K3
AI TOOL COMPARISON
Kimi K3 topped the public coding-arena leaderboard the day it launched, but a same-prompt, parallel test against Claude Fable 5 across four real build tasks tells a more complicated story.
01Kimi K3 model overview & specs
- Kimi K3 launched with roughly 2.8 trillion parameters and a 1 million-token context window, positioning it as a frontier-scale model.
- Its weights are scheduled for open release, meaning it can eventually run outside Moonshot's paid API once published.
02Kimi K3 vs Claude Fable 5 coding benchmarks
- Kimi K3 debuted at #1 on a public head-to-head coding arena leaderboard, ranking above Claude and GPT-class models for front-end code specifically.
- A leaderboard rank is a crowd-voted comparison score, not a guarantee of real-world coding reliability — the rest of the test measures that gap directly.
03Moonshot AI pricing & signing up
- Moonshot AI's entry tier for Kimi K3 access costs $19/month, comparable to a single mid-tier Claude subscription.
04Kimi Code vs Claude Code CLI interface
- Kimi's coding CLI ships a --yolo flag that lets the agent execute file writes and commands without per-action approval, the same trust model as Claude Code's autonomous modes.
05Coding challenge 1: playable voxel sandbox game
- Running identical prompts on both tools in parallel terminal windows let response speed and first-draft quality be compared side by side rather than sequentially.
- Claude Fable 5 sat in a visible 'thinking' state for close to three minutes with no output, while Kimi K3 began streaming visible reasoning almost immediately.
06Coding challenge 2: AI motion graphics video
- The motion-graphics prompt asked for a fully coded (not templated) piece — animated text easing, a soundtrack cue sheet, one runnable output file — a heavier deliverable than the other three tasks.
07Coding challenge 3: synth pad & drum sequencer
- The synth-pad prompt was included specifically to test polish and interaction design, not just functional correctness.
08Coding challenge 4: AI web design & live search test
- The website prompt withheld brand information and required the model to search the web for facts about the requester, testing live tool-use accuracy rather than pure code generation.
09Claude Fable 5 voxel sandbox game demo
- Claude Fable 5's voxel game shipped with working block-breaking, a day/night cycle, and Minecraft-comparable playability from a single prompt.
10Moonshot AI API usage limits & error issues
- The $19 Moonshot plan's five-hour usage allowance was fully consumed within minutes of running the four parallel prompts, forcing an immediate top-up mid-test.
- After the top-up, Kimi K3 returned repeated 404 'model not found' and permission-denied errors, attributed to server overload rather than a billing problem.
11Kimi K3 voxel sandbox game gameplay review
- Kimi K3's voxel game was judged visually denser and more graphically complex than Fable 5's, including a working sunset-to-night lighting transition — the one task where Kimi came out ahead.
12Claude Fable 5 Neon synth pad UI demo
- Fable 5's synth pad shipped a fuller control set (adjustable attack/release, a randomize function, multiple pad banks) than the equivalent Kimi build.
13Kimi K3 synthesizer UI comparison
- Kimi K3's synthesizer replicated the same core interaction pattern as Fable 5's but with fewer exposed controls, despite using more screen space.
14Claude Fable 5 web design results
- Fable 5's researched website correctly pulled the requester's real community platform, YouTube channel, and a separate business (an audio production company) from one unguided prompt with zero brand info supplied.
15Kimi K3 web design & search accuracy review
- Kimi K3's website was accurate on the same researched facts but visually busier and less on-brand than Fable 5's simpler layout — factual accuracy and design restraint are separate axes to judge on.
16Claude Fable 5 AI-generated video results
- Fable 5 delivered a fully rendered, on-brief motion-graphics launch video, complete with a synced soundtrack, from one prompt in under ten minutes.
17Kimi K3 AI-generated video comparison
- Kimi K3's equivalent video rendered with no audio track and simpler animation, with text bleeding outside its bounding box in places, and took noticeably longer to finish.
18Final verdict
- Across four identical, parallel-run tasks, Fable 5 won three of four outright and lost only the voxel game; a leaderboard rank did not translate into the reviewer's real spending decision.
- The reviewer explicitly separated 'topping a public leaderboard' from 'being worth paying for,' concluding that hands-on reliability outweighed the raw benchmark win.
Terms worth knowing.
- Arena leaderboard
- A public ranking site where AI models are compared head-to-head and voted on by users, producing a rank that reflects crowd preference rather than a controlled reliability test.
- YOLO mode
- A coding-agent setting that lets the AI execute file writes and terminal commands without approving each individual action, trading oversight for speed.
- Voxel sandbox
- A block-based 3D game environment, in the style of Minecraft, commonly used as a benchmark task to test whether an AI can build a complete playable game from a single prompt.
- Open weights
- Model files released publicly so anyone can download and run the model on their own hardware, instead of only accessing it through a paid hosted API.
Things they pointed at.
00:33productMoonshot AI ($19/month plan)
00:54toolKimi Code CLI (--yolo mode)
14:29toolGPT 5.6 (OpenAI)
Lines you could clip.
00:00
“A new Chinese model just beat Claude. This is not a press release.”
04:22
“Literally, in just a few minutes with $19, I hit a usage limit and all my projects have stopped.”
12:06
“Fable five has nailed it. It's got the brief spot on. That video was created with one prompt in literally under ten minutes.”
14:29
“So while Kimmy k three initially hits the top of the arena leaderboard, I'm still gonna be pumping my money into Anthropic and Fable five.”
Word for word.
Read-along
Don't just watch it. Burn it in.
See every word as it's spoken — crank it to 2× and still catch all of it. The same dual-channel trick behind Amazon's Kindle + Audible.
A new Chinese model just beat Claude. This is not a press release. This is a live leaderboard with Kimi k three right at the top above Claude Fable five and GPT 5.6.
So Kimi k three dropped yesterday and went straight to number one for front end code. It's got 2,800,000,000,000 parameters, a million token context, open weights this month coming, and you can use it right now.
So I'm gonna give it my hardest real world tasks, the jobs I normally save for Fable five, and some of the results really shocked me. Okay.
First of all, I'm gonna get signed up to use this new Open Frontier model, which is from Moonshot AI. I just have to scroll down to find their monthly plans.
Look at this. And I'll go for the very basic $19 a month plan to give it a test.
Okay. I've signed up for the $19 a month plan. Kimi k three is now fully ready to create things.
Okay. With Kimmy code installed, I just run Kimmy dash dash yolo to go into yolo mode and it will fire up right here. Look at this.
Fantastic. Let's run it again. Kimmy dash dash yolo and then now we've got two Kimmies on the right hand side, and we've got two Claude codes with Fable five on the left hand side.
Time for our first test. And I'm gonna use exactly the same prompt, build a playable voxel sandbox game. We'll get that running inside Fable five and inside Kimmy code so we can see exactly what happens, how fast they go, and which one is the winner.
Okay. We can see over here that Claw code has been thinking for almost one minute and thirty five seconds, and it's done. Absolutely nothing.
Whereas Kimmy k three is already getting to work thinking things through and showing its thinking, which I like a lot more than just this yellow canoodling. Uh, it's actually doing something and I can see that it's thinking it through. So we'll come back in just a moment and see how those do.
Now, while we're working away, Claude code still thinking for nearly three minutes with nothing happening here. Kimmy k three is definitely doing a little bit more there. I'm gonna paste in this prompt down below, create a motion graphics product launch video, entirely encode animated text easing, soundtrack cue sheet, output one file I can open.
And look at this, we've still got Claude Code thinking away for eight minutes saying that it's building it. It's already figured out how to make that Voxel game. And we're also thinking over here with Kimi Code.
Yeah. That is definitely still thinking right now what it's going to do, making all its plans. Down here, we've got the wonderful video, and we can see Python libraries are getting installed to render using Pillow drawing every single frame.
And over here, we've got Kimi k three thinking things through still and really working hard on this. Lots of thinking, no action yet, so let's trigger a few more projects so we can have quite a few things to look at by the end of this video. Okay.
We've got a brand new workspace, and here I'm gonna build a synth pad with some drum sequencing. It's gonna be really cool with Neon UI. That's running in Fable five and also now it's running in Kimi k three in parallel, which is great.
We'll see what the results are soon. And for the final prompt down here, I'm gonna ask it to design a premium website for CreatorMagic by Mike Russell, beautiful with Parallax Hero and Space for a Blog. And I'm asking it to search online to do this.
So we'll get Kimmy k three and Fable five running at the same time, building something. This will be a great test of design principles and also the model's ability to search online for information, bearing in mind I've told it nothing about Creator Magic.
Okay. We can see Fable five is often thinking, and already Kimmy k three is using its web search tool to find out about Mike Russell and Creator Magic, which is absolutely awesome. Despite Claude Code taking a lot of time to think with Fable five, we've already got the voxel game and this beautiful video completed and ready, which is amazing.
So we've got two things ready to view from Fable five. So let's play Fable five Voxelcraft. And, oh my goodness, as I would expect, this is actually really good.
There is nothing to say that this should be bad because, well, we know that Fable five can do an incredible job, and this is just incredible. Yes. I can even cut blocks.
It's basically like real Minecraft and the night and day has been followed exactly as expected even if a little bit fast. This is pretty decent. Fable five quality, exactly what we would expect.
Okay. Here's the really sad thing. You'll see that on my $19 that I put in on these projects, I've hit a four zero three on all of them.
I've hit my usage limit. Literally, in just a few minutes with $19, I hit a usage limit and all my projects have stopped.
So while I'm blown away and impressed with Fable five, I have to pump more money into Kimmy k three to actually get results. Let's hope that everything I've done so far isn't lost, and I can just pay them more money and hopefully get some results to show you.
Yep. Look at that. I literally blasted through my limit in just a few minutes here.
While I've got more seven day usage, my five hour usage has completely gone. Time to pay Moonshot AI more money. Thank goodness for Apple Pay and we're processing the money to Moonshot.
Oh, this is gonna be moonshot money and we're done. Back over in the terminal, continue and continue. We've got a four zero four not found the model or permission denied.
This is not great. Come on. Continue.
Please find the model and continue. Okay. Yes.
Now we're up and running. It's back and running. So let's do the same over here in this workspace.
Continue and continue. And hopefully, pretty soon, we're gonna have some Kimi k three results. Pretty expensive results, but we'll have them.
Now the funny thing is I keep getting four zero four not found on the k three model or permission tonight. This is extremely frustrating. I don't know if this is the service I should expect, especially as I keep pumping money into Moonshot AI without many results.
Continue. Continue. Continue.
Come on. Continue. Okay.
Good news. We've got another game up and running. This is Voxel Craft from, yes, you guessed it, Kimmy k three, something else to look at.
Let's play and, well, I've gotta say, it's pretty impressive. And in my opinion, this might actually be better than Fable five. So Fable five did something really well, but this definitely has more graphics.
Oh, my goodness me. Now, I guess I can change to build things as well if I want. Yes, with bricks.
Oh, my goodness. I mean, this is pretty insane. Is it gonna go from day to night, uh, just like Fable five did?
We don't quite know. This is good. Definitely more complex, but definitely awesome.
Yeah. I'm loving Kimmy k three's take on this. No nighttime yet.
We might need to wait a little while. And yes, look at this. The sun is actually going down.
I think very soon, we're experiencing sunset and, yes, dusk and night.
This is awesome. So both models do well. Although, in this case, I would say that Kimmy k three has just taken the edge over Fable five.
Well, seeing as if we've got this synth ready from Fable five, let's play Neon Pad. Oh, this is this is epic. Oh, I feel like a proper DJ.
Wow. This is amazing. Alright.
Let's put our hands up in the air and do some jamming. Oh my goodness. It actually works.
Hang on. Let's stop and see what we can do. We've got different synth pads.
This is insane. And I can trigger this with numbers. Oh, wow.
Okay. This is this is just absolutely awesome. So I can literally place anything I want on the pad.
I wonder if I can do random. Yes, I can. That was very random.
Let's do random. Oh my goodness. Let's clear everything and we'll do this.
This as a heartbeat like that. That sounds good. Okay.
Let's get that through. I'm having too much fun with this. Nice.
And then we can put little claps in between. Let's play.
I mean, it's messy, you know. I'm not a professional here, but I can actually play things on an instrument and change things. Wow.
This is amazing drum.
Oh my goodness. I can literally change everything about the signs being created.
Fable five is absolutely blowing me away. I wanna see some stuff from Kimi k three. So let's give this a go with headphones on.
This is the Kimi k three synthesizer. Now it takes up more real estate on the screen. Can I play?
Yes. Oh. Okay.
Pretty random and pretty similar, to be honest, to Fable five, if not a bit bigger. Let's try the number keys. In fact, it feels exactly the same.
Maybe a few less controls, I've gotta say. Let's try and bring some of this down, some of the attack and release down, change the shape.
Okay. This is cool.
Okay. That's all good. But, uh, what can we do with this?
Can we actually change this? Let's see if we can change this to just one beat here and then play that all back. So, we'll play.
Okay. And we'll bring the reverb down a little bit and put in a beat, a consistent beat maybe. I mean, it works.
It's good. Maybe a few less controls versus the Claude Fable five version. I would have to say, really, when it comes down to synths with drum pads, it's gotta be Fable five here.
K three pretty much replicated the same thing, just with less controls. Back over here inside the coding terminal, I'm still getting a lot of API errors. I can only imagine that things are basically overloaded with Kimi k three and everyone trying it right now, but it shouldn't be that way.
It should not be that way for a paying customer. Okay. Kimmy k three is still working away, whereas Claude Code has shipped a website, and this is for Creator Magic.
Ship real AI like magic. Lots of AI kind of creations here with the text spacing here. I'll kinda let that one slip.
The magic theme is definitely on brand. It's not quite my brand colors, but it's close. It's got a little git pull tank over here, which I think is really cool.
8,000,000 views on AI builds. Join the community. Actually links to my proper school community.
Look at this. If I click it, it will take me right there. Yep.
Into school. Watching on YouTube, if I click this, this takes me to, yes, my real YouTube channel. So I'm impressed with how much Fable five has cracked this.
Now, didn't get the logo, but I didn't expect that. Let's scroll down. Yes.
We've got all accurate stats here. Pick the skill you want.
Radio guidance AI builder. How funny is this? It researched this and got it right about me.
Uh, so this is absolutely fantastic and accurate. A place here for blogs exactly as I asked and subscribe to the newsletter. Let's click through.
Okay. This scrolls all the way around the website, which is nice. Uh, blog takes me to an actual blog page, which is great.
This is exactly what I asked for. And now look at this. Wow.
Make magic with AI. Creator Magic by Mike Russell. Weekly builds, AI agents, no code.
Yep. Absolutely. It's got my community link and my YouTube link absolutely correct.
This website is way better than I think what Fable five created. Definitely much busier. That may or may not be a good thing.
Okay. We've got Trusted By and some big names there, which is fantastic. Let's scroll down.
A lot going on here. This is maybe busier than I wanted, but again, it's pulling things in and it's quite interesting. It's mixed stuff here with my audio production company, 4,000 plus clients produced for.
That's my audio production company, Music Radio Creative. Ways to learn the craft. Oh, look at this.
Now, this is a holding space for my photo. I would imagine this is nice. Uh, again, we've got blog post areas here and the email list.
We've got About Mike as well. It scrolls around nicely to the blog, to the newsletter, to the community. While this is nice, I don't know.
I kinda liked the simplicity of Fable five despite Kimmy k three going more complex and cosmic in its design. Okay.
Now, we're just waiting for that video to be produced and it's really getting there. The final render is happening. Tank dash launch m p four is being made.
Now while Kimmy k three is still rock and rolling, yes, it's taking a lot longer, but maybe the results will be better. We've already got a launch video for tank. This is so exciting.
Let's watch the finished result.
Oh my goodness me. That was incredible. Fable five has nailed it.
It's got the brief spot on. That video was created with one prompt in literally under ten minutes. Okay.
Here's the great news. It's ready. Let's play the video and watch Kimmy k three's product launch.
Okay. No audio. I'm noticing that, but that's okay.
Yeah. It's pretty simplistic tank, create a magic, very nice, one agent, one terminal. You, babysitting, no.
Run your agents like field units. Not too bad. I kinda wish that there was some kind of audio like we got with Fable five.
This is all very nice. But, uh, yeah.
Where's the audio? Let's scrub through this and try and see the rest. Okay.
This is all very, uh, oh, that's actually a nice animation. Let's play this. Here we go.
Yes. We've got running. Nice.
That was a very good clock animation. That's a little bit complex and I don't like the fact that the words have kind of bled outside the box there. This is giving the tech stack.
Very nice indeed. Kind of basic though.
I definitely preferred Fable five. So, as we can see, in some cases, Kimmy k three took the lead. I think the voxel game, Kimmy k three was in the lead.
But personally, for me, out of three of the four tasks, I believe Fable five was in the lead and faster and included in my plan and I didn't hit API errors and, well, do I need to go on?
So while Kimmy k three initially hits the top of the arena leaderboard, I'm still gonna be pumping my money into Anthropic and Fable five. And who knows?
I might also have a bit of a play with GPT 5.6 Soul from OpenAI too. Let me know what you think of this video. Was this good?
Was I biased in any way? I don't think so, but you let me know in the comments down below. Thank you so much for watching, and YouTube is showing a video on your screen now.
You should watch next. Thanks.
The Hook
The bait, then the rug-pull.
A model called Kimi K3 launched and immediately took the #1 spot on a public coding-arena leaderboard, ahead of Claude Fable 5 and GPT-class models. So the host paid Moonshot AI $19 and ran four identical build prompts against Fable 5 in parallel, in real time, to see whether the leaderboard rank held up outside a vote count.
Named ideas worth stealing.
01:12list
Four identical parallel build tests
- Playable voxel sandbox game
- Coded motion-graphics launch video
- Synth pad with drum sequencer
- Researched brand website
The same four prompts were fired at both tools at the same time so speed and quality could be judged directly, not from memory across separate sessions.
Steal forAny tool-vs-tool review — run identical prompts in parallel, not sequentially, to remove recency bias.
How they asked for the click.
VERBAL ASK
14:40next-video
“Let me know what you think of this video... let me know in the comments down below... and YouTube is showing a video on your screen now you should watch next.”
Soft engagement ask (comment prompt) followed by a handoff to YouTube's suggested next video — no hard subscribe or product pitch.
MENTIONED ON CAMERA
FROM THE DESCRIPTION
PRIMARY CTAWhere the creator wants you to go next.
Visual structure at a glance.
