Kimi K3 vs Fable 5. Same prompt and reference. 3D globe dashboard. Kimi left, Fable right. To me, Kimi's UI fidelity is a touch better. The globe looks better too, with richer textures and smoother motion. This task: ~$3 with Kimi, ~$15 with Fable. I'd pick Kimi k3 for... frontend work.show more

@hqmank totally agree. the extra $12 for Fable isn't worth it when Kimi looks better and costs less. easy choice.

Honestly, I use Fable 5 less for coding. I find it more valuable for critical decisions and high-level thinking.

@hqmank interesting! what kinds of decisions do you use it for?

I usually tell Fable 5 not to write any code. I mainly use it for system architecture design, planning, and critical decision-making tasks, as well as fixing serious bugs. (These tasks already include the key decisions I mentioned earlier.)

@hqmank makes sense, keeping it focused. how’s it been with fixing serious bugs?

Besides safety-related cases, around 80% of the cases I run into might get routed to Opus because of the safety guardrails. For the rest of the tasks, I think it works pretty well.

@hqmank interesting breakdown! do you think Opus will adjust those safety guardrails soon?

If a task gets flagged by Claude's safety guardrails, I usually just switch to another model like Grok or Kimi K3 to get it done.

Same, tested Kimi K3 myself and honestly really impressed.

Cool, the frontend design consistency is really good.

thank u man

Kimi has been my frontend pick for a while now. Better result here and 5x cheaper is pretty hard to argue with.

Yeah, the only problem now is getting a Kimi coding plan 😄

Kimi K3 vs GPT-5.6 Sol. Side-by-side comparison here:

Where do you even start to build something like that ?

You can check my Twitter. I already shared the prompt and reference images there.

Checking !!!!!

That's a pretty compelling comparison. If you're getting comparable or even better frontend quality at a fraction of the cost, it's hard to ignore. That's exactly the kind of competition the ecosystem needs.Personally, I'd still keep both available. I use @kilocode , so I can reach for Kimi when I'm building UI and switch to another model when the task calls for it. I'd rather choose based on the project than lock myself into a single model.

@kilocode Exactly, that is basically how I use it too. I pick the model based on the task.

Looks like K3 has officially caught up with Fable 3. Absolutely stunning performance.

Kimi K3 has a real price advantage, and personally I prefer its frontend design quality. The visual taste and attention to detail feel really impressive.

AI competition is getting real, cost and quality both matter.

They are opening the model at the end of the month. I think more platforms will offer K3 API access for everyone then.

i choose kimi overall, my favorite ai

For frontend work, I would pick Kimi too.

good job, kimi now is 🐐 in frontend design

Yeah, no doubt. Kimi K3 is #1 on DesignArena's Frontend Web App Arena now.

Kimi k3 is pretty decent at front end indeed

Yeah, frontend is definitely where Kimi K3 shines. I will test its backend ability today.

Honestly, I can smell the AI slop coming from the right side, while the left seems to have masked it pretty well.

I agree. The left one has better earth texture and feels more realistic.

$3 versus $15 for similar output makes Kimi a hard one to argue against for frontend work.

Yeah, exactly. The coding plan is hard to get right now. If you want to try it, their API works too.

Might just test the API directly instead of waiting on the plan pricing.

Kimi K3 seems pretty good to me

People around me are saying the same thing. Too bad the Kimi coding plan is sold out right now 😄

What harness do you use?

This demo was built with Kimi Code.

Which backend the best?

I will start testing the backend part today. Stay tuned.

Kimi one looks way good

The Earth texture feels a bit more realistic. Kimi is ahead on this one.

this doesn’t prove anything, you have to test it on out of distribution tasks

This task probably is not in the benchmark distribution, so I think it is actually a better test of a model's real-world ability.

Kimi better

Same here.

Kimi 🔥

Yeah, it is #1 on DesignArena's Frontend Web App Arena right now.

I used Kimi models for frontend since I guess it was K2.5. K3 still have that edge, I'm not using it for any backend work - it can do it, but it takes longer and costs more in my experience compared to GPT-5.6.

Curious, why are you not using K3 for backend? Did you run into any issues? I am planning to test Kimi K3 on some backend tasks recently.

It’s slower and the price is higher than 5.6-sol in my experiences. It’s definitely capable of solving it. It’s a great model.

Thanks for the feedback! Yeah, the price is a bit high, and it is sold out at the moment.

I don’t think the price is unreasonable at all. It’s just that sol is really good value for money in the current landscape. And currently I can’t use it for my job. As I’m not sure they support zero day retention and not training on the the data.

Data security is also one of my main concerns. They probably need an official statement or clear policy around data retention and training before more people can use it for work.

I trust the @opencode team will get that done. I’m not sure if their current deal has this guarantee with the k3 model. But they do on the other models.

🚨News

K3 tops Frontend Arena on human preference votes, not automated tests. That's the benchmark that actually reflects what ships — and it's the one Fable loses on.

the 5x price gap is the real story tbh. if the cheap model is this close on ui one shots the default flips, you only reach for fable when the task needs the extra reasoning. been routing easy tasks to cheaper models for months, barely notice

Same here. Pretty much the same workflow. Use cheaper models for simple tasks, save the stronger ones for harder stuff.

Kimi wins the value battle easily

IMO Kimi wins the value battle easily. For frontend design, it’s the best right now.

@claudeai is being shot in the foot by @AnthropicAI . Something has got to give; the top 5% of users are going to leave. I've already done it—I'm never going back to that stupid Max subscription. People worry about privacy—okay bozo. @ollama for the win, baby.

Oh man

Bro, I suggest you give it a try if you get the chance. But it looks like they are sold out recently...

Oh kimi is much better. Seem like Kimi k3 really out-perform other models in frontend design

Super cool！ Can you share the prompt？ thanks！

here

The useful number is cost to an accepted result. If Kimi also needs fewer cleanup passes, the gap is much bigger than $3 vs $15.

What’s the prompt you used for this?

It's cheaper and still better.

Prompt?

Kimi's looks better

Comparing Kimi and Fable is a distraction. Models evolve every six months. Great architecture adapts. Build systems that outlive the models powering them.