The answers to “A category page from a product feed”, side by side

Playground · Business · HTML

A category page from a product feed

AI models build a rain jacket category page from a 12-product feed. We test the brand, colour, size and price filters and both sorts.

About this test

A shop category page is ordinary work, which is what makes it a good test: twelve rain jackets from a small product feed, with filters for brand, colour, size and price and sorting by price and rating. Nothing here is hard to draw. What matters is accuracy. Every price and rating has to match the feed, every filter has to return exactly the right jackets, and the page should not invent selling points the feed never mentions. We ran each page with scripts: filtering by Ridgeback, navy and XS, setting a price limit, sorting both ways, and reading the products shown after each step.

Read more

The prompt, word for word

Here are the products in an online shop's "Rain jackets" category, from its product feed (CSV):

id,title,brand,price_gbp,colour,sizes,rating,reviews
RJ-01,Packable rain shell,Northline,49.00,navy,S|M|L|XL,4.4,212
RJ-02,Packable rain shell,Northline,49.00,yellow,S|M|L,4.4,212
RJ-03,Storm parka,Fellwater,189.00,black,M|L|XL,4.7,96
RJ-04,Storm parka,Fellwater,189.00,olive,S|M|L,4.7,96
RJ-05,City trench,Arden & Vale,145.00,stone,XS|S|M|L,4.2,58
RJ-06,Lightweight anorak,Northline,65.00,red,S|M|L|XL,3.9,141
RJ-07,Trail jacket 2.5L,Ridgeback,119.00,blue,S|M|L|XL|XXL,4.5,330
RJ-08,Trail jacket 2.5L,Ridgeback,119.00,black,S|M|L|XL|XXL,4.5,330
RJ-09,Kids' puddle jacket,Little Fell,32.00,pink,3-4Y|5-6Y|7-8Y,4.8,74
RJ-10,Kids' puddle jacket,Little Fell,32.00,green,3-4Y|5-6Y|7-8Y,4.8,74
RJ-11,Insulated rain coat,Fellwater,159.00,navy,S|M|L,4.1,39
RJ-12,Cycling rain cape,Ridgeback,55.00,orange,One size,3.6,22

Build the category page for these 12 products: a product grid, filters for brand, colour, size and price, and sorting by price and by rating. Filters and sorting must work. Show every product with its correct price and rating; draw simple product images yourself with inline SVG or CSS.

Make it a single self-contained HTML file (inline CSS and JavaScript, no external libraries, fonts or images).
Show the whole prompt

Default settings: no thinking effort set, no temperature, and room for 64,000 tokens. A card says so when an answer was made with other settings: “limit 128k” means it could use up to 128,000.

Every answer runs live, in a sandboxed frame. Click into one to play with it.

How each model did

RankModelChecksTimeCost
1 Claude Opus 5.5 Our pick 4 of 4 2.5 min $0.361
2 GPT-6 Luna Best value 4 of 4 46 s $0.003
3 Claude Haiku 5.5 4 of 4 46 s $0.006
4 DeepSeek V4.1 Flash 4 of 4 117 s $0.013
5 Mercury 3 4 of 4 107 s $0.020
6 MiniMax M3.1 Flash (free route) 4 of 4 5.0 min $0.024
7 GPT-6.1 Terra 4 of 4 58 s $0.026
8 DeepSeek V4.1 Flash 1017 4 of 4 104 s $0.036
9 GPT-6.1 Sol 4 of 4 95 s $0.069
10 Claude Fable 5.2 4 of 4 5.7 min $1.479
11 Kimi K3.1 Pro 3.5 of 4 10 min $0.241
12 Kimi K3 3.5 of 4 8.1 min $0.457
13 Claude Fable 5.1 3.5 of 4 2.9 min $0.816
14 Muse Spark 1.3 3 of 4 3.3 min $0.045
15 Gemini 3.8 Flash 3 of 4 74 s $0.054
16 Gemini 4 Carbon 3 of 4 3.6 min $0.206
17 Qwen3.8 Max (0902) 3 of 4 18 min $0.246
18 Mistral Large 4 3 of 4 45 min2 tries $0.294
19 Muse Ember Pending 19 s $0.007
20 Gemini 4 Argon Pending 76 s $0.133
21 Grok 4.7 0 of 4 5.0 min5 tries $0.172

Ranked by our pick, then checks passed, then cost. Time and cost count every try; each dot is one check, in the order listed below (filled: passed, half: partly, empty: failed). Click a model for its notes.

What each answer cost, against the checks it passed

Best value: GPT-6 Luna, 4 of 4 checks for $0.003. Our pick: Claude Opus 5.5, 4 of 4 for $0.36. The best value is the cheapest answer with the most checks passed, counting every try.

0%20%40%60%80%100% $0.001$0.01$0.1$1$10 Cost, every try (log scale) Checks passed Claude Opus 5.5 GPT-6 Luna

Best valueOur pickOther answers

Notes on each answer

1

Claude Opus 5.5

Our pick 4 of 4

A clean four-column grid with a different drawing for each garment: a belted trench, a pouch-front anorak, a quilted coat and a flared cape. It offers checkbox filters with counts, min and max price boxes plus quick bands ("Under £50" correctly shows the four jackets at £32 and £49), and "Also in yellow" links between colours. Every filter and sort we ran was correct, and it adds nothing that is not in the feed.

2.5 min · 17,916 output tokens (3,902 thinking) · $0.361

2

GPT-6 Luna

4 of 4

"North & Field": a calm shop with breadcrumbs, checkbox filters for brand, colour and size, min and max price boxes, a "Clear all" link and four sorts. Every price and rating matches the feed, and every filter, combination and sort we ran returned exactly the right jackets; an impossible pairing shows "No jackets found". It adds only a tagline and an intro line, with no claims on the products. Every jacket is drawn as the same short-sleeved shirt, and the colour and size lists sit in short scrolling boxes, so red, stone, yellow and the children's sizes stay hidden until you scroll them. The cheapest of the first twelve answers, at $0.0033.

46 s · 6,556 output tokens (1,154 thinking) · $0.003

Added 8 October 2026.
3

Claude Haiku 5.5

4 of 4

A plain "Rain jackets" page with checkbox filters for brand, colour (with swatches) and size, min and max price boxes, a reset button and five sorts. Every filter, combination and sort we ran returned exactly the right jackets, and an impossible pairing shows "0 of 12 products". Its only addition is a header line ("Waterproof shells, parkas and rain gear for every season"), with no claims on the products. The trench gets a belt and the cape its own shape, but the other jackets share one short-sleeved outline. The cheapest of the first eleven answers, at $0.0056.

46 s · 11,022 output tokens (3,492 thinking) · $0.006

Added 8 October 2026.
4

DeepSeek V4.1 Flash

4 of 4

"Rainroom": a plain, well-spaced three-column grid with a two-handled price slider, checkbox filters and rain streaks behind each jacket. Every filter and sort was exact, and it adds nothing beyond the feed. It cost about a penny, the cheapest of the first ten answers by a wide margin.

117 s · 23,415 output tokens (15,545 thinking) · $0.013

5

Mercury 3

4 of 4

"Rain jackets" with a breadcrumb, a short "Waterproof styles for every forecast" line, a filter panel (brand, colour and size checkboxes, four price bands and "Clear all"), five sorts and a "Showing 12 of 12" count. Every price and rating matches the feed, and every filter, combination and sort we ran returned exactly the right jackets; Little Fell over £150 shows "No products match your selected filters." Every jacket is drawn as the same hooded shape in its colour, the cape included, and the cards do not name the colour, so the pairs read as duplicates. There are no min and max price boxes. 107 seconds, $0.020.

107 s · 57,274 output tokens (51,468 thinking) · $0.020

Added 19 October 2026.
6

MiniMax M3.1 Flash (free route)

4 of 4

"Rain & Ridge": a teal hero with a drawn umbrella and the counts (12 styles, 5 brands, 10 colours), then a grid of twelve drawn jackets, each tagged with its colour, with sizes, star ratings and review counts. Brand and colour checkboxes with counts, size boxes, four price bands and four sorts all returned exactly the right jackets in our tests, and an impossible pairing (Little Fell over £150) shows "No jackets match those filters". Every price and rating matches the feed. All twelve jackets are the same hooded shape in different colours, so the trench and the cape look like the parkas, and the hero line about "quick-drying shells" says a little more than the feed does. One try, 5.0 minutes on OpenRouter's free route, so the time is not representative; $0.024 at list prices.

5.0 min · 39,343 output tokens (23,244 thinking) · $0.024

Added 19 October 2026.
7

GPT-6.1 Terra

4 of 4

"The rain edit.", a clean three-column shop with dropdowns for brand, colour and size and minimum and maximum price boxes. All twelve jackets show the right prices and ratings, sizes are on every card, and every filter and sort we tried gave the right jackets, with "No products found" when nothing matches. The copy stays close to the feed. The drawings are tidy, but most jackets share one hooded shape. 58 seconds, $0.026.

58 s · 5,148 output tokens (141 thinking) · $0.026

Added 27 October 2026.
8

DeepSeek V4.1 Flash 1017

4 of 4

"Field & Fell": a four-column grid of twelve drawn jackets with their colours tagged, where the trench gets a belt and the cape its own flared shape. Brand and colour checkboxes with counts, size chips, four price bands and four sorts all returned exactly the right jackets in our tests, and an impossible pairing shows "No rain jackets found". Every price and rating matches the feed. There are no min and max price boxes, and the only additions are an intro line and a "12 waterproof styles" tag, which says a little more than the feed does. 104 seconds, $0.036.

104 s · 29,948 output tokens (20,162 thinking) · $0.036

Added 19 October 2026.
9

GPT-6.1 Sol

4 of 4

A quiet "Field & Form" shop with a serif heading, checkbox filters, min and max price boxes and large drawn jackets in cards. Every filter and both sorts returned exactly the right jackets in our tests, and it adds only a tagline, no product claims. The cards are big, so at desktop width you see two or three jackets at a time.

95 s · 6,820 output tokens (247 thinking) · $0.069

10

Claude Fable 5.2

4 of 4

A four-column grid with a different drawing for each garment (a belted trench, a pouch-front anorak, a spotted kids' jacket in a puddle), checkbox filters with counts, size chips grouped into adults, kids and one size, and min and max price boxes. Every filter and sort we ran returned exactly the right jackets, alone and combined. Options that would leave nothing are greyed out with a count of 0, so the page cannot be filtered down to an empty grid, and a price typed outside the range is pulled back to £32 or £189. It adds nothing that is not in the feed. As accurate as Opus 5.5's page, without its quick price bands or links between colours. 5.7 minutes and $1.48, the dearest of the first nineteen answers on this page.

5.7 min · 36,834 output tokens (21,896 thinking) · $1.479

Added 10 November 2026.
11

Kimi K3.1 Pro

3.5 of 4

"Field & Form", with counted tick boxes for brand, colour swatches, a size menu and four price bands. All twelve jackets are right, every filter and sort worked, and a reset appears when nothing matches. It draws the garments apart well: a belted trench, a pouch anorak, a quilted coat and a cape. Sizes appear only in the filter, not on the cards, and the intro promises "insulated parkas", which the feed does not. The judge saw only the start of the code and marked it down; the page works. 10 minutes, $0.24.

  • Adds no products, prices or claims that are not in the feed (partly)

10 min · 23,963 output tokens (14,855 thinking) · $0.241

Added 27 October 2026.
12

Kimi K3

3.5 of 4

A neat three-column grid where brand, colour, size and both sorts all work correctly. The price filter is broken: every band shows a count of 0, and ticking any of them, even "£150 and over", empties the page to "Showing 0 of 12". A footer calls the products, prices and ratings "illustrative".

  • Brand, colour, size and price filters work (partly)

8.1 min · 30,461 output tokens (22,509 thinking) · $0.457

13

Claude Fable 5.1

3.5 of 4

"WetWeather.co": a tidy three-column grid with checkbox filters, colour counts and a price slider paired with number boxes. All the filters and sorts gave the right jackets. It does add "Bestseller" badges to the Ridgeback trail jackets, a sales claim the feed does not make. It was also the most expensive of the first eighteen answers by far.

  • Adds no products, prices or claims that are not in the feed (partly)

2.9 min · 16,185 output tokens (3,928 thinking) · $0.816

14

Muse Spark 1.3

3 of 4

A plain shop with counted checkboxes, including four price bands that each return the right jackets, and each product's feed ID printed under its card. Every filter and sort was correct. It adds "Free delivery over £75 • 60-day returns" to the header, and the jackets look more like short-sleeved shirts.

  • Adds no products, prices or claims that are not in the feed

3.3 min · 10,393 output tokens (4,067 thinking) · $0.045

15

Gemini 3.8 Flash

3 of 4

"Aerostorm" has colour swatches, size chips, a single max-price slider and an "Add" button on every card. The filters and sorts all worked. But it stamps "20K WATERPROOF" on six jackets, the city trench among them, adds a free-delivery banner, and draws every jacket with the same short-sleeved shape.

  • Adds no products, prices or claims that are not in the feed

74 s · 14,167 output tokens (1,892 thinking) · $0.054

16

Gemini 4 Carbon

3 of 4

A clean white and teal shop with drawn jackets, all twelve feed entries and the correct prices and ratings. Brand, navy colour, XXL size and maximum-price filters worked, as did both directions of price and rating sort. It adds a free-delivery policy and a lifetime seam guarantee that the feed never supplied. 1 try, 3.6 min, $0.2056.

  • Adds no products, prices or claims that are not in the feed

3.6 min · 20,437 output tokens (1,551 thinking) · $0.206

Added 11 November 2026.
17

Qwen3.8 Max (0902)

3 of 4

The most styled page: "Downpour Supply Co." with a scrolling ticker, a big yellow-jacket hero, chip filters, price bands and a slider. Every filter and sort we tried worked. It surrounds the products with invented claims, though: a "10K MM / seam sealed" stamp, a "60-day storm test", "Back in stock" news for the Ridgeback jacket and a made-up weather widget. It took over 18 minutes.

  • Adds no products, prices or claims that are not in the feed

18 min · 40,797 output tokens (23,923 thinking) · $0.246

18

Mistral Large 4

3 of 4

An editorial "Stormline Supply" page with a large jacket hero, swatches, size chips with counts, a price slider and a basket. The filters and sorts all worked correctly. But it invents product facts: a "WATERPROOF 10,000 mm" badge, "taped seams and breathable membranes", and "New" and "Packable" tags, plus delivery terms and "Designed in Glasgow". Its first try was cut off at 64,000 tokens, and the second took 17 minutes.

  • Adds no products, prices or claims that are not in the feed

17 min · 76,327 output tokens (45,856 thinking) · $0.160 for this answer; 45 min and $0.294 over 2 tries

Took 2 tries: ran out of room at 64,000 tokens; finished at 128,000 tokens.
19

Muse Ember

19 s · 13,403 output tokens (6,816 thinking) · $0.007

Added 9 November 2026.
20

Gemini 4 Argon

76 s · 13,135 output tokens · $0.133

Added 13 October 2026.
21

Grok 4.7

0 of 4

Grok sent no code on its first three tries, and the fourth failed with an API error. The fifth try produced a "Bracken Outfitters" page, but its script stops with an error as it loads. The filter headings are empty, no jackets appear, and it still says "Showing 12 of 12".

Failed every check.

4.7 min · 26,555 output tokens (15,485 thinking) · $0.160 for this answer; 5.0 min and $0.172 over 5 tries

Took 5 tries: sent no code at 64,000 tokens; sent no code at 64,000 tokens; sent no code at 128,000 tokens; failed at 128,000 tokens, thinking off; finished at 64,000 tokens.Billed for 1,764 input tokens, about 1,179 more than the other models for the same prompt. We sent no system prompt, and the extra is there on every one of its requests, so it is text the provider adds itself.

One answer per model, at each model's default settings unless the card says otherwise, through OpenRouter. If an answer ran out of room or stopped short, we asked again with more room, then with thinking off, and say so. This is a quick look, not a benchmark: run it again and every answer will differ. For measured results, see our benchmarks.

How we pick. Claude Opus 5.5 tried every answer (clicking, dragging and playing it in a browser), marked the checks and drafted the notes. Our pick is a judgement among the answers with the most checks passed, not a score: the checks show what works, the pick is which of those we would rather use. Opus is an Anthropic model, and Claude answers are our pick in 18 of 29 tests, so weigh Claude picks with that in mind. Our method calls for a blind second opinion from another lab's model on any Claude pick; none is recorded for this page yet.

What we checked, by trying each answer ourselves: Shows all 12 products with the right prices and ratings; Brand, colour, size and price filters work; Sorting by price and by rating is correct; Adds no products, prices or claims that are not in the feed. A partial pass counts as half.

Verdict

Most models got the hard part right. All but two of the pages show all twelve jackets with the right prices and ratings, and almost every filter and sort did what it should. The differences are in restraint and finish. Claude Opus 5.5 is the best page: every check passes, it has the fullest price filtering (min and max boxes plus bands), and of the first ten it is the only one that draws a trench, an anorak and a cape as different garments (2.5 minutes, $0.36). DeepSeek V4.1 Flash and GPT-6.1 Sol also pass everything; DeepSeek V4.1 Flash did it for $0.013 in 117 s, and Claude Haiku 5.5 and GPT-6 Luna, added on 8 October, for under a cent. DeepSeek V4.1 Flash 1017, added on 19 October, passes everything too, with its own shapes for the trench and the cape ($0.036). So does MiniMax M3.1 Flash, added the same day on OpenRouter's free route, though all its jackets share one shape, and so does Mercury 3, added the same day ($0.020). GPT-6.1 Terra, added on 27 October, passes everything for $0.026, with one hooded shape for most jackets. Kimi K3.1 Pro, added the same day, has every filter and sort right and draws each garment differently, but its intro promises insulated parkas ($0.24). Claude Fable 5.2, added on 10 November, passes everything as well, draws each garment differently and greys out filter options that would leave no jackets, but it took 5.7 minutes and cost $1.48, four times what Opus 5.5's page cost.

Gemini, Qwen and Mistral built working pages but added waterproof ratings and other specs that the feed never gives, which a real shop could not publish. Kimi K3's price filter matches nothing at all. Grok 4.7 only produced code on its fifth try, and that page crashes on load and shows no products.

What would your prompt get?The Spring Prompt playground runs one prompt on several models at once, from OpenAI, Anthropic, Google, Mistral and others, side by side.
Ask for access
More from the playground
All tests →
Export
Include

Chrome will ask to share this tab: choose Allow. Only the board below is recorded.

Spring Promptspringprompt.com