Playground · Animation · HTML · Standing test: we add each new model as it launches
Holographic trading cards: frontier models, one prompt
A holographic card that leans towards your finger and catches the light is a small test of taste as much as code: a 3D tilt that feels physical, a foil that shifts with the angle without drowning the picture, a back worth turning over, and art drawn from scratch. We asked for a set of three for an imaginary collection, "Spring Prompt: First Editions", each card a creature the model invents. Every model got the same prompt at its default settings, with room for 64,000 tokens; where one ran out of room we gave it more, and say so. The answers below are exactly what they wrote, running in sandboxed frames. Move over each card, leave it alone for a moment, click it, then go to the next one.
Default settings here means we set no thinking effort and no temperature, and each answer could use up to 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.
The prompt, word for word
Design a set of three holographic collectible cards for a collection called "Spring Prompt: First Editions". Each card shows a small creature you invent and draw yourself (in CSS or SVG), its name, a line of flavour text and its number in the set (01/03, 02/03, 03/03). - The card in view tilts in 3D towards the pointer or finger, with a rainbow foil sheen and a glare that move with the tilt, and a shadow that shifts to match. - Left alone, it drifts gently on its own, so the foil keeps catching the light. - A click or tap flips the card over to show its back, which is designed too. - The viewer can move between the three cards. Make it a single self-contained HTML file (inline CSS and JavaScript, no external libraries, fonts or images). It should feel premium, fill the viewport, and work with both mouse and touch.
Every answer runs live, in a sandboxed frame. Click into one to play with it.
How each model did
| Model | Our notes | Checks | Time | Output tokens | Cost |
|---|---|---|---|---|---|
| GPT-6.1 Sol Our pick | Three pale pearlescent cards on a deep green page ("For the quietly curious"): Mossmallow, a leaf-eared bunny, Dewdrifter, a snail with a raindrop shell, and Petalpip. The foil is a soft sheen that turns from butter yellow to lilac as the card tilts (about 12 degrees each way), rather than a loud rainbow, and the line drawings and serif type stay crisp under it. Left alone it sways only a few degrees. The back is a dark foil card with a sprout emblem ("Collected, not captured"). The most finished design here, in about three minutes for $0.12. | 5 of 5 | 3.2 min | 11,927 1,034 thinking |
$0.120 |
| Claude Opus 5.5 | The closest to a real trading card: a carousel of three framed cards (Mossbun, Petalmoth, Puddlewhisk) with name bars, a type line ("Meadow sprite, HT 7 cm, WT 40 g") and flavour text. The frame changes colour as it tilts, from blue to green to red, and a sparkle layer runs over it; the back is a navy card with an engraved seal. The glare is strong enough to wash out the creature in the middle of the card. $1.04 and eight minutes. | 5 of 5 | 8.2 min | 51,727 31,185 thinking |
$1.036 |
| Claude Fable 5.1 | The loudest foil: a rainbow diamond pattern over the whole card that swings from green to magenta to gold as it tilts, with Mossling, a sprout blob, in front and two cards waiting either side. The creatures are cute and the back has a quiet sprout seal. It is more glitter than shine, and at $2.87 and 11 minutes the dearest answer by far. | 5 of 5 | 11 min | 57,380 39,626 thinking |
$2.872 |
| Claude Haiku 5.5 | Glassy violet cards (Thimblewisp, a moth on a tulip; Sproutback Tortle; Larkhollow Fawnkin) with a rainbow band that sweeps across as the card tilts, and a striking back: a rainbow crosshatch around a sprout medallion. The drawings are simpler than the best here, but every part of the brief works, with numbered tabs and a Flip button as well as a tap. $0.025: only DeepSeek V4.1 Flash passed every check for less. | 5 of 5 | 3.5 min | 50,038 35,210 thinking |
$0.025 |
| GPT-6 Luna | Muted olive cards from "Fieldnotes of the Fae": Mossling, a sprout with a heart, and Pipkin, who carries a lantern. Nicely drawn and set in a calm serif, and the card tilts and sways, but the foil is so faint that it barely reads as holographic. The back is a short written note. The fastest and cheapest answer, at $0.004 in 49 seconds.
|
4.5 of 5 | 49 s | 7,871 313 thinking |
$0.004 |
| Gemini 3.8 Flash | Dark cards with small creatures (Vernisprout, a dew gecko; Bloomveil, a squirrel) and the widest swing here, around 20 degrees when left alone. The foil is drawn as hard-edged blocks of gold, teal and purple that cut across the card and the text as it moves, so it looks broken rather than shiny. The back is a green foil card with a star seal. 96 seconds, $0.06.
|
4.5 of 5 | 96 s | 16,338 2,503 thinking |
$0.061 |
| Muse Spark 1.3 | The most game-like set: bright white cards with HP badges, element tags ("Leaf", "Mist"), rarity and rainbow foil edges, with Bloombit, Dewdrake and Pollenpuff drawn as round SVG characters. The tilt and foil work well; the back is a smaller affair, a dark card with an "SP" badge and the creature's description. Cheerful rather than premium. $0.05 in under three minutes. | 5 of 5 | 2.7 min | 11,811 3,021 thinking |
$0.050 |
| Grok 4.7 | An elegant idea, "A collector's folio" of three plates (Vellum Moth, Brookglass Newt, Firstbud Hare) in pastel pearl foil. But the foil is so pale and opaque that the names and flavour text are hard to read, the tilt is small (about 5 degrees), and on a phone the card covers the title. Its first try thought for 64,000 tokens and sent no code; the second finished. The two tries took 27 minutes and cost $0.65.Took 2 tries: sent no code at 64,000 tokens; finished at 64,000 tokens.Billed for 1,438 input tokens, about 1,171 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.
|
4 of 5 | 13 min 27 min over 2 tries |
59,018 42,254 thinking |
$0.355 $0.647 over 2 tries |
| Kimi K3 | Cream cards with gold frames (Bloomwhisk, a sprout bunny; Dewsprite, a raindrop; Petalflare), tidy and readable, with a back of concentric rings around a painted egg. The sheen is a single hard-edged diagonal band that slides across rather than a foil, so it reads as a glass reflection. It ran out of room at 64,000 tokens; the second try, with 128,000, finished. The two tries took 24 minutes and cost $0.41.Took 2 tries: stopped mid-page at 64,000 tokens; finished at 128,000 tokens. | 5 of 5 | 6.1 min 24 min over 2 tries |
31,332 17,633 thinking |
$0.408 $0.408 over 2 tries |
| Qwen3.8 Max (0902) | The foil is lovely on its own, a rainbow wash with a dotted pattern and a bright glare that follows the pointer, but it is laid over the card so heavily that the creatures (Dewlet, Flutterbud, Mosskit) cannot be seen, only their names. Clicking the card makes it disappear: the back never shows, and a second click brings the front back. 15 minutes, $0.22.
|
3 of 5 | 15 min | 36,432 20,803 thinking |
$0.219 |
| DeepSeek V4.1 Flash Best value | Dark glassy cards with a teal, purple and gold foil and a dotted grid, stacked as a fan (Petalwisp, a winged seed; Mossback Nib, a beetle; Sunmoth). It tilts, drifts and swipes well; the creatures are pale on dark and the small text is faint, and the back is a ring around the card number. A good answer for under a cent ($0.009), but it took eight minutes. | 5 of 5 | 8.4 min | 49,278 28,016 thinking |
$0.009 |
| Mistral Large 4 | Pastel holo cards with a "1st edition" stamp (Pokko, a moss sprout; Lumoth, a lantern moth; Ripple, a water spirit), named tabs below and a dark back with a gold "Spring Prompt" seal. The foil shifts from lilac to lime and the art is sweet, but the tilt is small (about 5 degrees), and a card you flipped stays flipped when you come back to it. It ran out of room at 64,000 tokens; the second try, with 128,000, finished. The two tries took 35 minutes and cost $0.32.Took 2 tries: ran out of room at 64,000 tokens; finished at 128,000 tokens. | 5 of 5 | 17 min 35 min over 2 tries |
87,331 56,041 thinking |
$0.183 $0.317 over 2 tries |
Time and cost are for the answer shown; when it took more than one try, the second line counts every try.
What each answer cost, against the checks it passed
Best value: DeepSeek V4.1 Flash, 5 of 5 checks for $0.009. Our pick: GPT-6.1 Sol, 5 of 5 for $0.12. The best value is the cheapest answer with the most checks passed, counting every try.
Best valueOur pickOther answers
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 9 of 19 tests, so weigh Claude picks with that in mind. GPT-6.1 Sol, shown anonymised screenshots of seven answers that passed every check (its own among them, unlabelled), ranked GPT-6.1 Sol's set first, Claude Opus 5.5 second and Claude Haiku 5.5 third. We agree with its pick.
What we checked, by trying each answer ourselves: Tilts towards the pointer, with foil and glare that move with it; Drifts on its own when left alone; Flips on a click or tap to a designed back; Moves between the three cards; The creatures are drawn and the card text is readable. A partial pass counts as half.
Verdict
GPT-6.1 Sol's set is the one we would print: pale pearlescent cards whose sheen turns from yellow to lilac as they tilt, with crisp line drawings and type that the foil never drowns, and a dark back worth turning over. It took about three minutes and cost $0.12. Claude Opus 5.5 made the most convincing trading cards, with colour-shifting frames, stats and an engraved back, but its glare washes out the art; Claude Fable 5.1 has the loudest foil, at $2.87 the dearest answer by far. Claude Haiku 5.5 passed every check for $0.025, and DeepSeek V4.1 Flash for under a cent.
The foil is where answers went wrong. Qwen3.8 Max laid it on so thick that its creatures vanish, and its cards disappear when flipped. Gemini 3.8 Flash's foil comes out as hard-edged blocks of colour, Grok 4.7's pale pearl makes its text hard to read, and GPT-6 Luna's is so faint that the card hardly looks holographic. This test gave each answer room for 64,000 tokens: Kimi K3 and Mistral Large 4 ran out and finished with 128,000, and Grok sent no code the first time.
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
