Pinned Tweet
Small Colosseum - big price difference 🔥
GPT-6.1 Terra vs Claude Opus 5.5 - building a pixel-art Roman arena with gladiators and spectators. Same task. $2.38 vs $8.38
Terra did a decent job for the price. But I prefer Opus here: the arena is brighter, has more depth, and feels more alive. It costs more, sure, but you can see the difference
Would you pay extra for Opus? 👀
GPT-6.1 Terra means business.
Our new fast, efficient model pairs sharper visual reasoning and precise computer use with clearer, more natural writing.
From decoding technical drawings to drafting proposals, filling forms to editing videos—it’s built for the whole workday.
GPT-6.1 Terra is rolling out today to Plus, Pro, Enterprise, and Edu users in ChatGPT Work and Codex.
NinjaCodex retweeted
Astra costs more and still loses on physics and colors 🦈
Claude Opus 5.5 vs GPT-6 Astra: jelly hammerhead shark
Opus nailed the jelly feel. The hammerhead bends and wobbles beautifully, and the colors actually look appetizing. Astra moves more like a worm than a jelly shark, the colors look washed out, and somehow it’s the more expensive one
Claude Opus 5.5 built this beach football scene for 9.29$
Real 3D with a pixel-art look: barefoot players, diving goalkeepers, waves washing onto the sand, palm trees and a beach bar. You can rotate the scene, change the wind, skim stones across the water and watch the beach light up at night
Prompt
Build a real-time pixel-art isometric diorama of beach football at sunset, as a single HTML page using three.js (WebGL). A floating island of sand with its own strip of sea held in by rocks; a taped-out pitch with two small goals and nets. Four-a-side, barefoot, two teams in bibs (red Sharks vs white Gulls), each with a keeper. Render at low resolution with an orthographic camera, snap colours to a fixed 64-colour OKLab palette, draw depth outlines, and add a reflection pass for the wet sea.
Simulate the ball physically: drag, wind, posts and crossbar, a net that catches it, and sand that kills the roll and the bounce. A ball in the sea floats back to shore on the swell; a ball off the edge brings in a spare from a fan.
Players with attributes (pace, power, accuracy, passing, control, tackling, flair) and simple AI. One chaser is chosen by intercept time; the others hold a formation. On the ball, a player chooses between solved lob passes to an open teammate, a flick up and volley, a chip over a keeper off his line, a hard drive, or an overhead kick. Add tackles that can leave the tackler face-down in the sand, and blocks.
Keepers read the shot from a forecast, dive, catch or parry, then hold and throw out. Kick-ins, corners and keeper's balls are fetched and taken properly (wading into the sea after the ball if needed). Goal celebrations; first to 5 or 7, then change ends.
Beach life: a bar with a string of lights, umbrellas and towels with fans who cheer and comment, a lifeguard tower, a fisherman on a jetty, a corn seller, a dog chasing dead balls, gulls, and a night fire with a guitarist. Waves wash up the sand with a foam lip. Day/sunset/night lighting.
UI: a scoreboard; a side rail with the team ladder, called moves (overhead/chip/long shot), weather and a live feed; click the sand to mark the next pass, click the sea to skim a stone. Original characters only, English UI, responsive down to phone width.
NinjaCodex retweeted
A desert oasis inside a jelly cube 🌴
Claude Opus 5.5 vs GPT-6 Astra - who did it better?
Opus nailed the soft sandy colors and beautiful water. And that sandstorm is so satisfying to watch - I just want to keep spinning the scene around
Astra has lots of little details to explore: camel caravans, garden irrigation, and buildings around the lake. The sandstorm looks a bit rough, though. I prefer Opus here
API cost:
Claude Opus 5.5 - $24.70
GPT-6 Astra - $32.11
Which oasis would you choose?
Opus 5.5 vs GPT-6 Astra. One prompt - two jousting tournaments ⚔️
Opus nailed the vibrant colors, lighting, and details. Astra created a nice atmosphere too, but visually, I’m giving this one to Opus
$14.65 vs $12.82. Less than two dollars apart. Who takes this round?
PROMPT
Make a real-time 3D pixel-art diorama (three.js, low-res render target, OKLab 64-colour palette, depth outlines, orthographic isometric camera that turns in quarter steps) of a medieval joust on a floating island. A sanded list with a wooden tilt hung with red-and-gold cloth, a low painted palisade, a three-tier grandstand with a royal box under a striped canopy, four round striped pavilions with pennons, lance racks, a marshal's stand, an armourer's lean-to, a pie stall, oaks, banners, torches and braziers. Four original knights with their own heraldry (field, ordinary, pixel charge) on shields, surcoats, caparisons and banners; plate armour, great helms with plumes. Horses as box rigs with two-bone IK legs and real gait footfalls (walk, trot, gallop) so hooves never slide. Each course: squires hand up lances, the herald sounds, the marshal drops his baton, the knights charge either side of the tilt, lower their lances and steer the point at the shield or helm with a skill-, gallop- and wind-based error; the hit point decides shield/helm/body/miss/foul, lance breaks into splinters, recoil, occasional unhorsing with a fall, a rearing riderless horse, a squire helping the knight up. Scoring 1 for shield, 2 for helm, unhorsing wins; semi-finals, final, the Queen hangs a wreath on the champion's lance. Crowd, ladies who throw favours when you click a knight, a juggling fool, crows, a dog. Controls: calls for the next course, slow motion at the strike, morning/noon/dusk, clear/windy/rain with mud and mirrored puddles. Scoreboard, bracket, knights list, feed from the stands.
NinjaCodex retweeted
What do you think of this jelly octopus? 🐙
Opus 5.5 showed once again who the jelly king is. Will Astra 6.1 and Fable 5.5 take the crown, or will the announcements be the only impressive thing about them?
Claude Opus 5.5 built this jelly tiger shark for $8.65 🦈
Interactive 3D with squishy physics: grab, stretch, and shake the shark to watch its jelly body wobble and bounce. You can even change its color and feed it little treats
Prompt
Build "Shark Jelly" — Material Studies No. 044: an interactive WebGPU soft-body study of a tiger shark cast in two layers of jelly, lying on its belly on a pale studio sweep. Same engine as the jelly series (WGSL, no libraries): signed-distance shape → surface-nets render mesh + snapped hex lattice (5 tets/cell) for XPBD with co-rotational shape matching, volume constraints, floor friction, soft grab with twist; shadow map, top-down contact AO, back-face thickness, refraction, Beer–Lambert absorption, thin-part glow, PBR Neutral tone map.
Shape: elliptical section swept along the spine — blunt, broad, almost square snout, deep chest, slim caudal peduncle, belly pressed flat. All fins from one recipe (flat triangle, convex leading edge, falcate trailing edge carved by the circle through its ends, thinning to the tip, smooth-unioned): tall first dorsal, small second dorsal, long pectorals, pelvics, anal fin, heterocercal tail with a long notched upper lobe and a short lower one that props the tail on the table.
Two layers: clear jelly over a milky belly whose top is the countershading line, dropping to the mouth corners. Shader markings in rest space: wavering, forking, broken tiger bars acting as a pigment film (spots toward the head), five gill slits, underslung mouth with labial furrows and a row of milky teeth, spiracles, nostrils, ampullae pores, lateral line; eyes with green-gold iris and slit pupil; trapped bubbles.
Interaction: "Swish tail" drives the body by muscle strain — each tet's rest shape is lengthened/shortened along the body by a travelling wave of curvature × lateral offset (volume-preserving), so the tail really beats and a firm shark creeps forward. "Drop bait": jelly beads that swell into being over the head, fall, bounce, collide with the deformed body and can be tossed. Palettes Steel / Amber / Gummy; readouts mass, volume %, fork length, live tail bend in degrees; firmness & damping sliders, ¼ speed, mesh overlay, pause, reset; editorial UI in English.
NinjaCodex retweeted
Opus, did you really have to bring the whole neighborhood to life? 🏀
Claude Opus 5.5 vs GPT-6.1 Sol. One prompt, two 3D pixel-art streetball courts - Sol also got a revision
Opus nailed the colors, lighting, and atmosphere. I could just sit and watch this little courtyard
Sol mostly pulled it off, but even after the revision, some issues remain: all six players celebrate every basket - including the team that conceded 😂
$10.44 vs $5.33. Almost twice the price. Would you pay the difference?
Badminton in the park, AI rivalry behind the scenes 🏸
GPT-6.1 Terra vs Claude Opus 5.5 - building a pixel-art badminton court with players and greenery
One prompt. $3.23 vs $10.73
Opus delivers again with rich detail, vibrant colors, and beautiful lighting. But Terra built a cozy scene too - for less than a third of the price
Would you pay extra for Opus? 👀
PROMPT
Build a real-time 3D pixel-art diorama "Sunday Net": one-on-one grass volleyball in a city park on a floating island (three.js 0.160, orthographic isometric camera with quarter-turns, low-res render target, depth outlines, 64-colour OKLab palette, hard shadow map, day/dusk/night). Park: 8×5 m grass court taped in white, worn sand patches at the serving spots, net top 2.30 m on two blue steel poles with guy ropes, white tapes and red-white antennae over the sidelines; a duck pond with stone bank, reeds, bulrushes, lily pads and a willow, its water rendered as a screen-space planar reflection (mirrored scene pass) with glints and expanding ripple rings; gravel paths with cast-iron lamp posts and benches, a yellow kvass barrel cart with parasol and seller, a picnic on a checked cloth, a striped players' blanket with bag and bottles, lindens, a bicycle, a round flower bed, a low clipped hedge round the edge. Ball physics: 21 cm, 270 g, quadratic drag, Magnus force for topspin jump serves, knuckling wobble for float serves, grass bounce and roll, net/tape collisions (rolls over or drops back), crossing outside the antennae is out, floats and drifts on the pond. Shot solver: for each launch angle bisect the speed to reach a target at a given height (ground across the net, or set height on one's own side), choose the angle by touch (pass/set apex, fastest spike that clears the tape, softest cut), iterate the aim against the wind as the player reads it, add skill/hurry noise and shanks. Rules: one player per side may touch the ball three times running — pass to self, set to self near the net, approach-jump-attack (spike, cut shot, deep shot) or a free ball over; rally scoring to 15 (or 21), win by two, winner of the rally serves, winner of the set stays on, four friends rotate from the blanket. AI predicts the flight and picks a contact family (forearm pass, side digs, overhead set, spike with jump height, dive), reads the attacker's position on defence, leaves balls going out (sometimes wrongly). Box-rig players with IK: platform pass, hands over the forehead, arm-swing jump spike, dive and push-up, toss and float/jump serve. Retrieval: pick up off the grass, wade into the pond knee-deep (ducks paddle away), spare ball from the bag. Mallard ducks that dabble and come to crumbs from a grandmother on the bench; a man walking a dog along the paths; a waiting player buys a mug of kvass. Fireflies by the pond, moths at the lamps, butterflies over the flowers; wind calm/breeze/gusty. UI: flip scoreboard (set, rally, ball speed), three-touch indicator, attack-call buttons (spike/cut/deep/jump serve), set-length switch, players ladder, feed of remarks, wind readout, view controls; clicks aim the next attack or throw bread to the ducks.
Claude Opus 5.5 claude.ai/artifact/Ri6JQERod…
GPT-6.1 Terra sunday-net.vercel.app/
NinjaCodex retweeted
Opus, and this is only $5.19? 🐬
Claude Opus 5.5 vs GPT-6.1 Sol. One prompt - two jelly dolphins
Opus surprised me again: vibrant colors, beautiful jelly, and satisfying physics. Usually it impresses me with quality, but this time with the price too
Sol at $2.92 fell short here: the physics weren’t quite what I wanted, and the colors lacked vibrancy
Just $2.27 more - I’m picking Opus. What about you? 👀
Hockey on the ice, AI rivalry behind the scenes 🏒🔥
GPT-6.1 Terra vs Claude Opus 5.5 - building a pixel-art winter hockey scene with players, snow-covered cabins, and a frozen rink. Same task. $2.98 vs $16.67
Opus delivered as always: detailed scenery, beautiful ice, and lighting that brings the whole scene to life. But Terra did a great job for the price too - a cozy winter world for almost 6x less
Would you pay extra for Opus? 👀
PROMPT
Build a real-time pixel-art isometric diorama of hockey on a frozen river, as a single HTML page using three.js (WebGL). A floating winter island: a frozen river crosses it from edge to edge and spills off both ends as icicle falls. On the black ice, a rink has been shovelled out and walled with a ring of snow lumps, with a gap for the bench and a small red-framed net with white mesh at each end. The nets are set in from the walls so players can go behind them. The north bank climbs to a terrace with snow-roofed log houses (smoking chimneys, icicles on the eaves, lit windows), a fence, birches with crows and a sledging slope. The south shore is low: firs, a snowman, a fire and a small bath-house. Add a footbridge upstream and an ice fisherman on a box over his hole downstream. Render at low resolution with an orthographic camera, snap colours to a fixed 64-colour OKLab palette, draw depth outlines, and reflect the scene in the ice.
Play three skaters and a goalie a side, village rules: no lines, no offside. The goalies wear felt boots and pads. Skaters have momentum: pushes to accelerate, long glides, limited turning, and hockey stops with a spray of snow. They skate backwards when defending and fall and slide on their backs after collisions. Simulate the puck physically: low friction on ice and high on snow, solved passes (v0² = v² + 2ad), saucer passes over sticks, lifted shots aimed through a spot in the net, soft snow walls that can bury a hard puck, posts and crossbar.
AI: a chaser chosen by intercept time, attack and defence shapes, a presser and markers, poke checks, dekes, wristers, snapshots, slapshots, backhands, wraparounds and one-timers, plus blocked shots. Goalies read the shot from a forecast, drop into the butterfly, stretch, glove or kick out rebounds, and cover the puck.
An old man drops the puck at faceoffs, catches the puck the beaten goalie slides up to him, and tosses spares when a puck is lost in the snow. A dog digs the lost pucks out and brings them back.
The rink is a live canvas: skate blades scratch it white, falling snow settles on it and slows the puck, kids clear lanes with snow scoops, and between games the old man floods it from buckets into a mirror.
Village life: sledders, the fisherman catching perch, a woman crossing the bridge with a yoke of buckets, fans on a bench who comment, a boy on the snow wall who throws snowballs on click, crows, chimney smoke, breath in the frost. Day/dusk/night lighting with lamps, a string of bulbs and the fire; clear/snow/blizzard weather.
UI: a scoreboard, a side rail with team ladders, called moves, weather, snow on the ice and spare pucks, and a live feed. Original characters only, English UI, responsive down to phone width.
try it here
GPT-6.1 Terra river-hockey.vercel.app/
Claude Opus 5.5 claude.ai/artifact/SymHoBwBb…




