# Outfits test — 2026-09 (round one, text only)

**Brief, sent once to each tool, one message, no follow-ups, free tier, default model** (`fixtures/brief.md`, which also holds the dress-code reference written before the test):

> Man, 35, average build, going to a wedding in Portugal in late September - outdoor ceremony at 4 pm, about 28 C, dress code 'garden formal'. I already own navy chinos, a white oxford shirt and brown loafers. Suggest 3 complete outfits that use what I own where sensible, each with the specific items to buy (brand, item name, link to an EU shop, price), total under EUR 250 per outfit. Tell me which one and why, and what 'garden formal' rules out.

Run by Claude Code in the operator's Chrome session on 2026-09-14 between 11:04 and 11:12 (+07), checks 11:15–11:50; the brief was typed into each tool rather than pasted; every output read and every check reviewed by the operator. Normal chats. Raw outputs are in `raw/`. Round two (a photo of the owned items) has not been run.

**Measured figure: item links right, of the links given.** Every item link was opened the same hour and counted right only if it lands on a product page (not a search, category, brand or home page) for the item named, in stock in at least some sizes, from a shop that ships to the EU, at a price within 10% of the page - a lower sale price counts as right. Dress-code sanity was checked against the reference: a jacket present, nothing named that the code rules out. Massimo Dutti's Spanish site would not render in our browser; two of Claude's links are counted unverifiable rather than wrong.

| | ChatGPT | Claude | Gemini | Perplexity | Copilot | DeepSeek | Grok | Qwen | Kimi |
|---|---|---|---|---|---|---|---|---|---|
| Date run | 2026-09-14 | 2026-09-14 | 2026-09-14 | 2026-09-14 | 2026-09-14 | 2026-09-14 | 2026-09-14 | 2026-09-14 | 2026-09-14 |
| Item links given → right | 6 → 1 | 4 → 0 (+2 unverifiable) | 11 → 0 | 6 → 2 | 5 → 2 | 0 | 2 → 0 | 9 → 0 | 7 → 1 |
| What was wrong | Amalfi beige blazer (its pick) and navy blazer **sold out in every size**; both ties €39.99 and €50, not €22.99; linen trousers link is a category | Zara and H&M links are category pages (the H&M one is trousers, offered for a shirt); Massimo Dutti ES pages would not render | **nine "links" are Google searches**; the other two are ASOS and Zara category pages | Zara tie and both trousers URLs redirect to search (product not found); pocket square is a US shop | Mango home page for the pocket square; Mango GB ties category; Massimo Dutti **US** polos category | no links at all | two category pages (Zara DE, H&M DE) | Mango ES old-format URLs redirect to the home page; Zalando and ASOS slugs have no product id; five home pages | six of seven links are shop or category pages |
| What was right | Mango green linen blazer €89.99, in stock | — | — | Zara PT linen blazer €89.95 in stock; ASOS Topman navy DB blazer in stock, ~€115 vs "€90–110" | Massimo Dutti IE linen blazer €160 (Copilot: ~€169), in stock - its pick; ASOS Mango Milan sage blazer ~€120, in stock | — | — | — | Zara PT linen blazer €89.95 in stock - its pick |
| Jacket in every outfit? | yes | **no** - outfit C is a camp-collar shirt and chinos; it calls this "the riskiest read on formal" | yes | yes (two outfits complete) | yes; outfit 3 is a knit polo under a blazer, which its own rule-out list calls out | yes | yes | yes | yes; outfit 2 is a knit polo under a blazer, flagged as "peak garden formal in 2026" |
| Rule-out list vs reference | matches; adds "no bridal white", "no boat shoes" | matches; adds "head-to-toe white" | matches; adds espadrilles | matches | matches | matches; adds "no all-black or all-white" | matches; adds "thin-soled oxfords on grass" | matches | matches; says ties are optional, "a jacket is effectively mandatory" |
| All three under €250? | yes (€113–163) | yes (€75–230) | yes (€222–230) | two priced (€133, €173); third cut off | yes (€180–240) | **no** - outfit 3's HUGO blazer is €355 | yes (ranges) | yes (€138–223) | yes (€165–185) |
| Uses the owned items? | all three; outfit 3 swaps the chinos | all three | 1 and 3 drop the chinos for a suit | yes | yes; outfit 3 drops the shirt | all three, all three items | yes | yes | yes; outfit 2 drops the shirt |
| Pick and reason given? | yes, #1 with five reasons and a caveat about the chinos' cut | yes, A | yes, 1 | **no - answer stops mid-outfit-3** | yes, 1 | yes, 1 | yes, 1 | yes, 2 | yes, 1 with three reasons and a fallback |
| Answer language | English | English | English | English | English | **Chinese** | English | English | English |
| Follow-ups to a shoppable list | 2 | 3 | 3 | 2 | 1 | 3 | 3 | 3 | 2 |
| Output form | prose with Mango product chips; ~1 min | prose; ~1 min | text with every item name linked; ~30 s | text with URLs written out, 104 sources, videos tab; cut off with a Pro upsell | text with URLs, three follow-up chips | markdown, no URLs | markdown, two category links | markdown with links, 78 sources | text with shop-level links after five searches; "prices checked today" |
| Quality / Effort / Cost / Free tier (0–3) | 1/3/3/3 | 1/2/3/3 | 1/2/3/3 | 1/2/3/2 | 2/3/3/3 | 1/1/3/3 | 1/2/3/3 | 1/2/3/3 | 2/3/3/3 |
| What stood out | The best styling advice and rule-out list of the nine, attached to a top pick that is sold out and two ties at double the quoted price. | Sends you to Zara's category page three times and offers an outfit with no jacket to a "formal" wedding. | Every item is a hyperlink and nine of them are Google searches. | The only tool whose links are written out - and half of them 404; it ran out of answer before the pick. | The one answer whose top pick opens on a product page, in stock, at the price stated. | Three blazer names, no links, one over budget, in Chinese. | Sensible outfits, two category links, every price a range. | Nine links, none of which lands on the product. | Honest about prices ("US listing shows $59.99, EU typically…"), one verified product link, six shop pages. |

## Notes per tool

### ChatGPT

Six Mango Portugal links. The green 100% linen blazer (37061391) is a product page, €89.99 down from €129.99, sizes 50–56 available. The Amalfi beige blazer - ChatGPT's top pick - is a product page at €36.99 down from €129.99 (ChatGPT quoted the list price) with every size marked "Não disponível"; the navy linen/lyocell blazer likewise, every size unavailable. The burgundy silk tie is €39.99 and the navy patterned silk tie €50.00 on their pages; ChatGPT gave €22.99 for both. The linen trousers link is Mango's linen-trousers category. The prose is the best of the set - a five-reason pick, "make sure the chinos are genuinely tailored-looking", loafers on grass, no bridal white - but you cannot buy the outfit it picked.

### Claude

One search. Outfit A's blazer, outfit B's jacket and trousers all link to the same Zara ES "linen suits" category page. The two Massimo Dutti ES product URLs exist but the page would not render in our browser at all (it hung the tab); counted unverifiable. The H&M "linen (IE)" link is the men's trousers category, offered for a linen camp-collar shirt. Belts and pocket squares have no links and "~" prices. Outfit C is a shirt and chinos with no jacket, which the code rules out; Claude says so itself. The pick, A, is sound in principle.

### Gemini

Every item name is a link. Nine go to google.com/search; the ASOS one is the men's suits category and the Zara one the accessories category. "BOGGI Milano / ASOS Design" names two brands for one blazer. The outfits are all jacketed and priced, and outfits 1 and 3 replace the owned chinos with a suit, which "where sensible" allows.

### Perplexity

Forty-five seconds, 104 sources, mostly a Hoi An tailor's blog. URLs are written out in full: the Zara PT 100% linen suit blazer (p04410410) is real, €89.95, in stock - the same product Kimi chose. The Zara tie (p04388740), suit trousers (p04410420) and relaxed linen trousers (p02736240) all redirect to Zara's search page: not found. The pocket square is bows-n-ties.com, a US shop, offered as EU. The ASOS Topman double-breasted linen blazer is a real product, in stock in most sizes, ~€115 against "€90–110". The answer stops mid-sentence in outfit 3 and never gives the pick the brief asked for; the free tier appended a Pro upsell.

### Copilot

Fourteen "actions". The Massimo Dutti Ireland 100% linen suit blazer (l02600356) is a product page, €160 (down from €200; Copilot said ~€169), add-to-basket available - and it is the top pick. The ASOS Mango Milan sage blazer is a product page, ~€120, in stock in sizes 46–50 and 56. The Mango pocket square links to Mango's home page, the tie to a GB category, and the knit polo to Massimo Dutti's US polos category. Outfit 3 puts a knit polo under the blazer after the rule-out list said "no polo shirt as the main top layer"; Copilot hedges that hosts may expect a collared shirt.

### DeepSeek

Thirty pages searched; the answer is in Chinese for the second time today. Each outfit keeps all three owned items and adds one blazer, named but not linked: a LOSILOO set at ~€74.95 that "seems to ship only to Ireland", an OVS Piombo at ~€129, and a HUGO Karlo at ~€355, over the budget, which it acknowledges. Ends by telling the user to search Zalando or ASOS for "Men's linen blazer beige". The reasoning for the pick - the light jacket balances the two "heavy" owned colours - is good.

### Grok

Three jacketed outfits, every price a range, two links, both categories (Zara DE linen blazers, H&M DE linen blazers), and generic item names ("Zara 100% Linen Suit Blazer"). The rule-out list is one of the better ones - thin-soled oxfords on soft grass, shiny fabrics that show sweat. "Shop EU sites for easy returns."

### Qwen

Seventy-eight sources and nine links, none of which opens on a product: the two Mango ES URLs are an old format that redirects to Mango's home page (one of them is the product id ChatGPT's green blazer used successfully in the PT format), the Zalando and ASOS links are slugs with the product id missing, and five are home pages. The outfits themselves are sound and jacketed, and the "spezzato" reasoning for outfit 1 is right.

### Kimi

Five searches, with visible reasoning about US-versus-EU prices. One product link - the Zara PT linen blazer, €89.95, in stock, its top pick - and six shop-level pages (Mango home, Scalpers, COS, Uniqlo, Mango Outlet, Brucle categories) with prices given as "~" and an explicit caveat. The pick has three reasons and a named fallback if the size is sold out. Outfit 2 is a knit polo under a Uniqlo blazer, argued as within the code. It also says "ties are optional" and "a jacket is effectively mandatory", which is what the references say.

---

# Round two — one photo of the owned items, 2026-09-15

**Brief, sent once to each tool in a fresh chat with `fixtures/owned-items.jpg` attached** (the photo brief and what the photo actually shows are in `fixtures/brief.md`):

> Same wedding as before: Portugal, late September, outdoor ceremony at 4 pm, about 28 C, dress code 'garden formal'. The photo shows what I already own. First tell me exactly what you see - each piece, its colour, fabric and any details. Then build one garden-formal outfit around these three pieces: what to add, with brand, item name, link to an EU shop and price, under EUR 250 total. Tell me if anything in the photo is a problem for garden formal.

Run by Claude Code in the operator's Chrome session on 2026-09-15 between 20:23 and 20:37 (+07), links checked 23:50–00:25 (+07) that night; the brief was typed into each tool (pasted into Grok, whose editor drops typed spaces); every output read and every check reviewed by the operator. Free tiers, default models; Claude incognito, Copilot temporary, Grok private; the rest normal chats. Fresh chats with no round-one context.

**Deviation.** The protocol asked for a photo of the operator's own wardrobe. The fixture is instead three Wikimedia Commons photos composed on one plain background (attributions in `fixtures/README.md`): stone pleated cotton chinos, a white poplin shirt with double cuffs and cufflinks in, and tan suede tassel loafers with a readable Tod's insole. So the photo deliberately does **not** match the round-one wardrobe (navy chinos, oxford shirt, brown loafers) - the test is whether a tool describes what is in front of it.

**Measured figure: pieces read right, of three.** A piece counts as right when the colour family and the garment type are both right. Details (pleats, double cuffs, suede, tassels, the brand on the insole) are recorded separately. Links were opened the same night and counted right by the round-one rule: a product page for the item named, in stock in some size, from a shop that ships to the EU, at a price within 10%. H&M GB would not open for us (bot wall, 403 to curl as well), so Claude's one link is unverifiable rather than wrong.

| | ChatGPT | Claude | Gemini | Perplexity | Copilot | DeepSeek | Grok | Qwen | Kimi |
|---|---|---|---|---|---|---|---|---|---|
| Pieces read right (of 3) | 3 | 3 | 3 | 3 | 3 | **1** - trousers only | 3 | 3 | 3 |
| Pleats seen | no ("flat front") | **yes** ("double front pleats") | no | no | "or very subtle pleating" | no ("no visible pleats") | no | no | **yes** ("складка спереди") |
| Double cuffs / cufflink seen | yes / yes | yes / yes ("shell or knot motif") | yes / yes | "button cuffs" and "decorative cufflinks" | yes / no | — | yes / yes | no / no | yes / yes |
| Suede · tassel · Tod's | yes · yes · yes | yes · yes · yes (+ gommino sole) | yes · yes · yes | yes · yes · "Tod's or similar" | yes · yes · yes | — | yes · yes · yes | yes · yes · no | yes · yes · yes |
| Invented details | — | — | "stark white", "cutaway" | breast pocket; button cuffs | — | — | — | — | turn-ups |
| Is anything a problem? (reference: nothing is) | no; shoes 4/10 | **the shirt** (too dressy) and the chinos | the chinos and the loafers | the trousers "main risk"; loafers "relaxed" | the trousers "weakest" | the trousers | **none** | shoes and trousers "too casual" | leather sole on grass; chinos "on the edge" |
| What it adds | navy MD linen blazer €199 + belt €39.95 = €238.95 | H&M beige linen-blend jacket ≈€100 + belt = €125–130; drop the shirt | Uniqlo navy blazer + cufflinks + square + knit tie + belt = €151.77 | Bexley linen blazer €129 + belt + socks + tie = €182–206 | Mango sky-blue linen jacket ~€160 + Pini Parma belt ~€49 = ~€209 | a shirt €159 + loafers €90 (+ belt) ≈ €249 | Zara beige linen blazer €90–100 + square €15 = €105–125 | H&M navy linen jacket €89.99 + Zara tie €49.95 + square = €150–155 | Mango ecru linen jacket €99.99 + Zara belt €25.95 + square + socks ≈ €141 |
| Jacket added? | yes | yes | yes | yes | yes | **no** | yes | yes | yes |
| Under €250? | yes | yes | yes | yes | yes | yes (but buys the wrong things) | yes | yes | yes |
| Links given → product pages → right | 2 → 2 → **0** (both Massimo Dutti pages "not found") | 1 → 1 → unverifiable (H&M GB blocked) | 0 (five domain names rendered as Google searches) | 2 → 0 (Bexley and ASOS categories) | 2 → 1 → 0 (Mango home; Pini Parma belt is £85, not ~€49) | 2 → 2 → 0 (Care of Carl shirt **sold out**, "€159" is the site's beanie promo; Nordstrom Rack is US) | 1 → 1 → **1** (Zara ES linen blazer €89.95, six sizes) | 2 → 0 (H&M IE and Zara MT categories) | 4 → 2 → **2** (Mango NL jacket €99.99, sizes 50/52; Zara ES belt €25.95) |
| Language | English | English | English | English | English | English | English | English | **Russian** |
| Output form | prose + two product chips; 43 s | prose; ~1 min | text with search "links"; ~20 s | table + prose, 16 sources; ~10 s | prose, 15 actions; ~1 min | table; ~40 s | prose, sources chip; 19 s | bullets with links; ~1.5 min | table + prose, 3 searches; ~2 min |
| Quality / Effort / Cost / Free tier (0–3) | 2/3/3/3 | 2/2/3/3 | 1/2/3/3 | 1/2/3/2 | 2/3/3/3 | 0/1/3/3 | 2/2/3/3 | 1/2/3/3 | 2/3/3/3 |
| What stood out | Read all three, kept all three, one navy blazer and a belt - then both links 404. | The best eye of the nine (pleats, gommino sole, cufflink motif), then tells you to leave the best shirt at home. | Read the Tod's logo; every "link" is a Google search. | Right on the big three, invents a breast pocket and button cuffs; two category links. | Right on all three, honest that the trousers are the weak piece; belt price about half the real one. | Saw one of three pieces and tried to sell you the other two, from a sold-out UK page and a US shop. | The only one to say the photo has no problem, and the only link that opens on the item, in stock, at the price. | Three thin lines, then "buy other trousers and shoes". | Three of three with the cufflink, two links that open at the price - in Russian, to an English brief. |

## Notes per tool

- **ChatGPT.** Massimo Dutti AT `l02008266` and PT `l06194660` both return the shop's "page not found" (checked 00:05). The descriptions are careful (says it cannot establish fibre content), the plan is one jacket and a belt, and it grades each piece as a "problem out of 10". Memory was on; nothing from round one surfaced.
- **Claude.** Incognito, Sonnet 5 Medium. Fullest read: pleats, metal button, French cuffs with the cufflink motif, Tod's "gommino" sole. Its one problem piece is the French-cuff shirt, which the reference does not rule out; it recommends a barrel-cuff shirt instead. H&M GB link unverifiable.
- **Gemini.** Flash. Reads the insole logo; no product links - the five "EU shop links" are bare domains with "Search '…'" instructions and the UI wraps them in google.com/search. Calls both the chinos and these loafers too casual.
- **Perplexity.** 16 sources, mostly a Lisbon tailor's blog. Right on the three pieces, then "button cuffs", "decorative cufflinks" and a "single visible breast pocket" that is not there. The Bexley €129 blazer has no product link; the two links are categories. Only tool to insist on a tie and socks.
- **Copilot.** Temporary chat. Three of three; "flat front or very subtle pleating" is the nearest to the pleats after Claude and Kimi. shop.mango.com is the home page; the Pini Parma page is real and the belt is as described, but £85.00 (≈€100), not "approx. €49 (listed by retailer)".
- **DeepSeek.** Search on. Describes only the trousers and never mentions the shirt or the shoes, then recommends buying a shirt and loafers. The Care of Carl page is sold out and shows "SHOP STENSTRÖMS FOR AT LEAST €159 – GET A BEANIE", which is where the "€159" came from. Nordstrom Rack ships from the US.
- **Grok.** Private chat. Three of three with the cufflink and the insole logo. "None of these items are a problem" - matches the reference written before the run. Zara ES `p04410410`: product page, €89.95, sizes 46–56 selectable. The rest are "search on asos.com / zalando.de" instructions.
- **Qwen.** Three one-line bullets, no cuffs, no brand. Says the shoes and trousers are too casual and offers to find dress trousers and oxfords instead. Both links are category pages.
- **Kimi.** Three of three, pleat and cufflink included, one invented turn-up. Both product links open at the price it gave: Mango NL Amalfi jacket €99.99 (down from €159.99, ecru, 50 and 52 in stock), Zara ES 100% leather belt €25.95. Total ≈€141, the cheapest complete plan with verified links. Answered in Russian because the account interface is Russian ("Language: ru per system" in its shown reasoning) - the brief and round one were in English.