Two models, one prompt-thread each, a minimalist photography portfolio with 3D CSS/JS touches. Part of our ongoing /design benchmark series in Command Code.
The prompt
Create an elegant minimalist single-file HTML photography portfolio landing page for "Elena Voss Photography". Tailwind CSS. Hero: full bleed hero image with 3D floating frame effect around the main photo (CSS 3D + JS). Infinite scroll masonry grid that loads more images with smooth fade-ins. Clicking an image opens a 3D lightbox gallery. About section with subtle parallax. Journal / Stories section with scroll horizontal cards. Contact with elegant form. Use soft elegant typography, lots of whitespace, and tasteful 3D interactions. Very high aesthetic quality. Output complete HTML.
Models and cost
| Model | LOC | Est. cost (floor) |
|---|---|---|
| Kimi K3 | 655 | ~$0.14 |
| Qwen-3.8-Max | 889 | ~$0.03 |
Kimi K3 sent 2 prompts in one thread (build + continue); Qwen-3.8-Max was a single agentic /design run. Both figures above are floors: final kept file's output bytes ÷ 4 × the model's output price - they only count the last file you'd actually keep, not the rest of the session.
Neither number is the real story here. Kimi K3's actual session cost is closer to ~$0.27 once you account for a long upfront design-planning reasoning pass before the single write of the whole file (a --continue turn only ran a syntax check, no rewrites). Qwen-3.8-Max's real session cost is unknown and the ~$0.03 floor likely understates it badly - this repo's one actually-metered Qwen-3.8-Max /design run, from the creative-portfolio benchmark, billed a real $6.70 for a similarly-sized file, driven almost entirely by call count and accumulated input context (35 API calls, 2.7M input tokens, mostly cache reads, plus screenshot-based verification) rather than output size.
What this comparison can and can't tell you
This benchmark has cost and line-count data for both models but hasn't been through a full craft scorecard yet - no collision-style code citations, no hands-on browser playtest of the 3D lightbox or masonry grid. What we can say: Qwen-3.8-Max wrote about 36% more code (889 vs. 655 lines) at a lower output-only floor, but given how badly Qwen-3.8-Max's floors have undershot its real metered costs elsewhere in this benchmark series, that floor comparison alone isn't a reliable basis for picking a winner on price.
Try this yourself
1npm i -g command-code1/model kimi-k3
2/design Create an elegant minimalist single-file HTML photography portfolio landing page. Tailwind CSS. Hero: full bleed hero image with a 3D floating frame effect. Infinite scroll masonry grid with smooth fade-ins. Clicking an image opens a 3D lightbox gallery. About section with subtle parallax. Journal section with horizontal scroll cards. Elegant contact form.Swap /model qwen-3.8-max to compare. Full command reference: commandcode.ai/docs/slash-commands/design. More one-shot comparisons: CommandCodeAI/slash-design-showcase.
