“Load post” shows the post from X only when you tap it. Until then, this page loads nothing from X.
Generations · Characters
3D Mario test: Astra vs every Sol A 3D Mario character across OpenAI generations: the author rates Astra best (15m 37s, $2.96) and GPT-6.1 Sol the best of the Sols (34m 38s, $0.64).
Costs: author-reported , likely API pricing. Not my receipts.
17 likes as of Oct 1 0:19 video
Showcase · Games worth a look
Chess drawn as crayon doodles on paper A playable chess game drawn as crayon doodles on paper, with cute creature pieces, built with Sonnet 5.5.
892 likes as of Sep 30 1:14 video
Showcase · Motion & video worth a look
Behind the 13,081-tile code mosaic Opus 5.5 made a 75-second behind-the-scenes video of a code-only, 13,081-tile goldfish-bowl mosaic, tracing one tile back through the code that drew it.
834 likes as of Sep 30 1:15 video
Showcase · 3D scenes worth a look
Photoreal koi pond built in the browser Sonnet 5.5 built a near-photoreal browser koi pond with lily pads and caustics, and recorded the real-time capture on its own.
536 likes as of Sep 30 0:10 video in Japanese
Showcase · Games
Poke the squishy jelly animals “poke.” is a squishy jelly-animal toy game, the author’s fourth game made with Opus 5.5 since its launch.
341 likes as of Sep 30 0:46 video
Other · 3D scenes worth a look
Fantasy shop diorama: GPT-6.1 Sol vs Claude GPT-6.1 Sol (High) built “Moonlit Curios”, a rainy isometric fantasy-shop diorama, in the ChatGPT app using 1% of the Plus weekly limit; the author says it has less impact than Claude’s, but is cute.
Reports usage as a share of a subscription limit. 332 likes as of Sep 30 0:38 video made in ChatGPT app (Work mode, Sites skill) prompt shared in Japanese
3+ models · 3D scenes
Felted Halloween figurines in 3D: three-model showdown Three models built 3D needle-felted Halloween figurines; the author says “Gemini 4 Pro” (an Arena checkpoint) won, Sonnet 5.5 did well and GPT-6.1 Sol fell short.
Uses a model name that has not been officially announced. 256 likes as of Sep 30 0:23 video
3+ models · 3D scenes
WoW Mulgore walk: four models compared A “WoW forever walk” test: four models built a third-person walk through a Mulgore-like savanna with MMO-style HUDs.
223 likes as of Sep 30 0:16 video
Head to head · Games
3D ski jump game: GPT-6.1 Sol vs Opus 5.5 GPT-6.1 Sol and Opus 5.5 each built a 3D ski-jump game from the same prompt and reasoning mode; both show full game HUDs.
142 likes as of Sep 30 0:30 video in Japanese
3+ models · Motion & video
Four models animate a fox running through the forest Four models animated a fox running through a forest from the same prompt; the author says Claude clearly leads and GPT-6.1 Sol can’t match Opus 5.5.
Opus 5.5 high Effort: high, level 3 of 5 Sonnet 5.5 high Effort: high, level 3 of 5 GPT-6.1 Sol max Effort: max, level 5 of 5 GPT-6 Sol max Effort: max, level 5 of 5 Models ran at different effort levels. 134 likes as of Sep 30 0:13 video
Showcase · Motion & video
Opus 5.5 motion design in three prompts Claims Opus 5.5 made a UI micro-interaction motion-design video in only 3 prompts, framed with the line “We no longer hire motion designers.”
130 likes as of Sep 30 0:10 video not a first try, says the author in Chinese
3+ models · 3D scenes
Four models build the Eiffel Tower in 3D A 2x2 of 3D Eiffel Towers from GPT-6.1 Sol, GPT-6 Astra, Sonnet 5.5 and Kimi K3; the author says GPT-6.1 Sol “isn’t that good”.
116 likes as of Sep 30 0:28 video
Head to head · 3D scenes
Night train viaduct: Sonnet 5.5 vs Opus 5.5 A “night train benchmark”: Sonnet 5.5 and Opus 5.5 each render a lit train crossing a stone viaduct at night; the author was impressed by Sonnet.
93 likes as of Sep 30 0:25 video
Head to head · 3D scenes
Red train through a low-poly city GPT-6.1 Sol vs GPT-6 Astra on the same prompt: a low-poly city with a red three-car train.
92 likes as of Sep 30 0:23 video
Head to head · Characters
Chrome bunny mascot: Sonnet 5.5 vs GPT-6.1 Sol Sonnet 5.5 and GPT-6.1 Sol each built an interactive chrome bunny mascot with a floppy ear; the author can’t pick a winner.
83 likes as of Sep 30 0:30 video made in TypingMind prompt shared
3+ models · 3D scenes
Crochet Halloween bouquet: four-model 3D showdown Four models built a 3D crocheted Halloween bouquet; the author found Sonnet 5.5 the most creative and GPT-6.1 Sol the best arranged.
Uses a model name that has not been officially announced. 71 likes as of Sep 30 0:28 video
Head to head · Apps & UI
Opus 5.5 vs Sonnet 5.5 build a reptile shop Opus 5.5 vs Sonnet 5.5 building a reptile-shop website: the author finds Sonnet uses space better, and Opus has nicer illustrations and clearer copy.
66 likes as of Sep 30 0:37 video in Japanese
Head to head · Physics
Mentos + Coke eruption: Opus vs Sol Opus 5.5 and GPT-6.1 Sol got the same prompt for an interactive 3D Mentos + Coke eruption and simulated the physics very differently.
65 likes as of Sep 30 1:00 video
Head to head · 3D scenes
Sonnet vs GPT-6.1 build a 3D apartment complex A 3D apartment complex from Sonnet 5.5 XHigh and GPT-6.1 Sol Ultra; the author says Sonnet won but cost about twice as much.
Models ran at different effort levels. 52 likes as of Sep 30 0:15 video
3+ models · 3D scenes
Four AI snow globes, one shot each Four models built a 3D snow globe from the same prompt in one shot: Opus 5.5 and Sonnet 5.5 (default effort) vs Sol 6.1 and Sol 6.0 (high).
Models ran at different effort levels. 51 likes as of Sep 30 0:15 video first try, says the author
Head to head · Physics
Jelly fruit under a hydraulic press A 3D jelly-fruit hydraulic press: Opus 5.5 got it right from the first prompt ($14.30); GPT-6 Astra needed two follow-ups ($32.88 in total).
Costs: author-reported , likely API pricing. Not my receipts.
Only one of the models got it on the first prompt. 51 likes as of Sep 30 0:18 video first try, says the author
Head to head · 3D scenes
3D car factory assembly line: GPT-6.1 Sol vs Sonnet 5.5 GPT-6.1 Sol and Sonnet 5.5, both at xhigh, each built a 3D car-factory assembly line; the author says 6.1 Sol is far better than GPT-6 Sol.
49 likes as of Sep 30 0:20 video in Korean
Cost compare · 3D scenes
Floating island railway: Opus 5.5 vs GPT-6.1 Sol Reshared comparison of a floating-island railway scene: Opus 5.5 ($14.84, 1 h 32 min) vs GPT-6.1 Sol ($7.15, 45 min), with Sol called the winner.
Costs: author-reported , likely API pricing. Not my receipts.
Original work by @dropout_layer . Also posted by @noclipepe .
49 likes as of Sep 30 0:45 video
Head to head · 3D scenes
GPT-6.1 Sol vs Opus 5.5: Eiffel Tower threejseval’s Eiffel Tower scene: Opus 5.5 High scored far higher Elo than GPT-6.1 Sol High but used ~5x the tokens and ~7x the cost ($1.15 vs $0.16).
Costs: author-reported , likely API pricing. Not my receipts.
Posted by a benchmark operator. 46 likes as of Sep 30 0:42 video made in threejseval (own harness)
Head to head · 3D scenes
Floating-island train world at half the price Floating islands with a train above the ocean: GPT-6.1 Sol (45 min, $7.15) scored 9.5/10 vs Opus 5.5 (92 min, $14.84) at 8.0.
Costs: author-reported , likely API pricing. Not my receipts.
Also posted by @OmNawale45831 .
43 likes as of Sep 30 0:52 video
Head to head · Motion & video
Blind test: AI-coded launch films with fuzzy mascots Blind test: GPT-6.1 Sol (XHigh) and Sonnet 5.5 (Medium) each coded an unofficial launch film for OpenAI’s “Dots” from the same brief; names are revealed at the end.
Blind test: costs were not attributed per model. Models ran at different effort levels. 34 likes as of Sep 30 1:00 video
Cost compare · 3D scenes
Pelican bike ride: three models, three price tags A 3D pelican-on-a-bike scene from three models with build costs: GPT-6.1 Sol $0.70, GPT-6 Astra $3.71, Opus 5.5 $3.54 (“delight per dollar”).
Costs: author-reported , likely API pricing. Not my receipts.
Posted by a company promoting its own product. 32 likes as of Sep 30 0:25 video made in AIHubMix (own gateway)
Cost compare · 3D scenes
Acrylic keyboard: $1.78 vs $57.34 Same Three.js acrylic-keyboard prompt in UltraCode: GPT-6.1 Sol took 11 min and $1.78; Sonnet 5.5 took 78 min and $57.34; the author says Sonnet’s has more detail.
Costs: author-reported , likely API pricing. Not my receipts.
29 likes as of Sep 30 0:30 video made in UltraCode
Head to head · 3D scenes
Pirate ship swing ride: Opus vs Sol A 3D pirate-ship swing ride by Opus 5.5 (high) and GPT-6.1 Sol (max); the author says Opus clearly wins.
Models ran at different effort levels. 28 likes as of Sep 30 0:22 video
Head to head · Motion & video
Commuter cyclist animation: Astra vs Sol GPT-6 Astra vs GPT-6.1 Sol on one prompt, code only, no fixes: a commuter cycling home at sunset. The author says Sol’s rider has backwards legs.
27 likes as of Sep 30 0:30 video first try, says the author in Japanese
Head to head · 3D scenes
Lego bricks assemble a toy car in 3D Sonnet 5.5 and GPT-6 Astra got the same short prompt: a 3D animation of Lego bricks assembling into a toy car.
25 likes as of Sep 30 0:16 video prompt shared
Showcase · Shaders & VFX
GPT-6.1 Sol's ClearWater ocean, now on phones GPT-6.1 Sol optimized the author’s real-time ocean shader demo, ClearWater 6.1, for phones with touch controls; a live link is included.
15 likes as of Sep 30 1:15 video not a first try, says the author
Head to head · 3D scenes
3D strawberry: Opus 5.5 vs GPT-6 Astra First-draft 3D strawberries from Opus 5.5 and GPT-6 Astra; the author says the Opus one would suit a farming game.
12 likes as of Sep 30 0:15 video first try, says the author
Showcase · Games worth a look
Opus 5.5 turns a vague idea into a broom-flying game A creator says Opus 5.5 turned a vague idea into a playable 3D broom-flying game almost instantly, and that anyone can do this.
782 likes as of Sep 30 0:15 video in Japanese
3+ models · Motion & video
Four models turn one line into animation TypingMind gave four models one line and asked each to animate whatever it imagined; the 2x2 video compares Sonnet 5.5, Opus 5.5, Fable 5.1 and GPT-6 Astra.
Posted by a company promoting its own product. 235 likes as of Sep 30 0:43 video made in TypingMind
Showcase · Explainers
Opus 5.5 turns a camera-move cheat sheet into a three.js lesson Opus 5.5 turned an infographic of ten AI-video camera moves into a ~5-minute three.js teaching video with BBC-style narration, after a “surprise me” prompt.
211 likes as of Sep 30 5:02 video prompt shared in Chinese
Cost compare · Physics worth a look
Wobbly watermelon jelly slice in WebGPU The official Three.js account shared a playable Claude artifact and @vib3coded’s video of “Melon Jelly”, a WebGPU soft-body watermelon you can grab and cut (Opus 5.5, $12.48).
Costs: author-reported , likely API pricing. Not my receipts.
Original work by @vib3coded .
5,014 likes as of Sep 30 0:18 video made in Claude Code (inferred) not a first try, says the author
Showcase · Motion & video
Sonnet 5.5 Max codes a whole video BridgeMind shares a 60-second video it says Sonnet 5.5 made entirely in code, and claims Sonnet 5.5 on Max effort beats Opus 5.5.
A claim without a side-by-side result. Posted by a benchmark operator. 970 likes as of Sep 30 1:00 video
Head to head · Physics worth a look
Stretchy gummy octopus: Opus vs Astra Opus 5.5 and GPT-6 Astra each built a stretchy gummy octopus (“Octo Jelly”); the author prefers Opus’s squishier version, which also cost less ($14.01 vs $20.74).
Costs: author-reported , likely API pricing. Not my receipts.
758 likes as of Sep 30 0:18 video
Cost compare · Physics
Jelly orange slice with a knife and heart cutters Opus 5.5 added a knife and shape cutters to a WebGPU jelly orange slice, and every cut piece keeps wobbling. Base $7.58 + update $11.60 = $19.18.
Costs: author-reported , likely API pricing. Not my receipts.
314 likes as of Sep 30 0:18 video not a first try, says the author
3+ models · 3D scenes
Steampunk Blender whale: Opus vs Sonnet vs GPT Atomic Chat had Opus 5.5, Sonnet 5.5 and GPT-6 Sol each build, texture and animate a steampunk whale in Blender; the three renders play side by side.
Posted by a company promoting its own product. 236 likes as of Sep 30 0:23 video made in Atomic Chat (own harness)
Head to head · 3D scenes worth a look
Sonnet 5.5 vs Astra: 3D three.js Jeep Same prompt and reference image for a single-file three.js Jeep: Sonnet 5.5 used 6% of the weekly limit (very detailed, slow); Astra used 7% (cleaner showcase page).
Reports usage as a share of a subscription limit. Models ran at different effort levels. 226 likes as of Sep 30 0:50 video
Showcase · Motion & video
Zooming out from Earth to the observable universe A cosmic zoom-out from Earth to the observable universe, built in the browser with Opus 5.5 (xhigh) over about 2 hours.
Opus 5.5 xhigh Effort: xhigh, level 4 of 5 170 likes as of Sep 30 0:35 video not a first try, says the author in Korean
Cost compare · Physics
$1.95 vs $4.29 gummy jellyfish showdown Sonnet 5.5 ($1.95) and GPT-6 Astra ($4.29) each built a stretchable WebGPU gummy jellyfish; the author describes Sonnet’s as more transparent and Astra’s as plumper and glossier.
Costs: author-reported , likely API pricing. Not my receipts.
119 likes as of Sep 30 0:18 video
Showcase · Motion & video
Opus 5.5's 16-hour shader-made bench nuke animation In a nearly 16-hour run, Opus 5.5 turned the idea “slumped on a bench while an atomic bomb explodes” into a code-made animation with sound, writing a shader per shot.
87 likes as of Sep 30 0:32 video in Chinese