Claude Fable 5.2 vs Opus 5.5 vs Astra 6: Opus Dethroned

Anthropic released Claude Fable 5.2 on 9 November at $8 per million input tokens and $40 per million output, a fifth less than Fable 5.1 and still twice the price of Claude Opus 5.5. Anthropic’s own benchmark table puts it ahead of Opus 5.5 almost everywhere, and Artificial Analysis ranks it first on its Intelligence Index at 65, six points clear of Opus 5.5. This series checks a claim like that a different way: hand the model the same build briefs as its rivals, allow one attempt, and publish whatever comes back without touching it.

Its opponents bring the builds they were published with. Claude Opus 5.5, the model Anthropic recommends for everyday work and the one that has led this series since September, built them for its 22 September round. GPT-6 Astra, OpenAI’s flagship, built them for its head-to-head with Claude Fable 5.1. The briefs: a motorway interchange seen from above and a seaside fairground at dusk, each as a single animated SVG file, and a cinematic rocket launch as one web page. One attempt each, maximum reasoning effort, and no opening a browser or running code to see the result.

Fable 5.2 wins with 14 points. Opus 5.5 takes 12 and Astra takes 9. Fable 5.2 took both SVG rounds with builds in which nothing is broken, and finished a point behind Opus 5.5 on the launch. It is the first time Opus 5.5 has finished anywhere but first in a test in this series, and Fable 5.2 was also quicker than Opus 5.5 on all three briefs.

Round 1: The interchange

The brief: a busy interchange seen from above, as one self-contained SVG with no text and no script. Two roads cross on different levels, ramps connect them, at least twelve vehicles follow the curves of the roads they drive on in both directions, and the layering hides the lower road’s traffic under the overpass. Two traps go unstated: whether each ramp really joins the roads at both ends, and whether traffic keeps to the right side after it merges.

Claude Fable 5.2 built a complete cloverleaf in daylight: two six-lane highways crossing at a bridge, four outer ramps for the right turns and four 270-degree loops for the left, the whole scene turned ten degrees off the grid so it reads as a place and not a diagram. Fifty vehicles drive on it, two more than Astra’s, among them vans, three articulated lorries and two buses. Around the roads there are striped fields, ponds inside two of the loops, a farm at the end of a dirt track, lane arrows painted on all four approaches and sign gantries over the lower road.

The traffic plan is the careful part. Through traffic keeps to the two inner lanes of each carriageway and the outer lane is left to cars using a ramp, so nothing ever has to merge into a gap. Measured across two full cycles, no two vehicles in the same lane come closer than about 60 pixels, bumper to bumper. A car turning left on a loop starts on one level and finishes on the other; Fable 5.2 draws each of those eight cars twice, once under the bridge deck and once over it, and swaps the copies halfway round the loop, on open ground where the change cannot be seen. Loop cars also ease down to half speed through the curve and pick it back up on the way out. Every animation in the file runs for 24, 12 or 8 seconds, so the scene repeats exactly every 24 seconds, and every vehicle starts and ends its run outside the frame.

Then there is the detail that has cost almost every model on this brief. Each highway’s solid white edge line stops at every ramp opening and becomes a short dashed line, the way a real road marks the mouth of a slip road, at all sixteen openings. The ramps carry edge lines of their own and meet the carriageways in long curves. We could not find a fault in it.

Claude Opus 5.5 planned its traffic down to the merge. In 58 minutes 49 seconds it built four right-turn ramps, one in each quadrant, each leaving the outer lane of one road and joining the outer lane of the other, with start times staggered so every merge lands in a gap, and a turn signal on every car that takes a ramp. Its bridge has parapets, expansion joints and a shadow on the road beneath, and its ramps peel away in smooth curves. Its fault is a painted line: each carriageway’s solid edge line runs straight across the point where a ramp leaves or joins it, at all eight ramp ends, so each ramp appears to slide out from under the road. It built no loops, so its interchange offers right turns only.

GPT-6 Astra set its interchange at blue hour and filled it, in 16 minutes 15 seconds. Forty-eight vehicles run along eight routes, each throwing a headlight cone onto the road ahead, with direction arrows painted on the asphalt, street lighting down the median, two ponds and a river. Its bridge is a real structure, with bearings at all four corners, and its occlusion works mid-crossing: a lorry enters with its cab already hidden under the parapet while its trailer is still in the open. Its road edge is drawn as two lines, and one of them is wrong: the thicker outer edge stops where each ramp leaves, while the thin white shoulder line inside it runs on through the exit.

Round 1 goes to Fable 5.2, which takes 5 points: it missed nothing, down to the dashed line at each exit. Astra takes 4 for the most atmospheric picture of the three, and Opus 5.5 takes 3.

Round 2: The fairground

A seaside fairground at dusk, one SVG. A Ferris wheel with at least ten gondolas turning continuously and staying level with the horizon all the way round, a spinning carousel with at least six horses rising and falling out of phase, twinkling bulbs, lights chasing the rim of the wheel, and the whole scene reflected in water. The gondolas are the trap the brief states out loud. The carousel is the one it does not.

Fable 5.2’s scene runs from a deep violet sky through pink to a low sun on the horizon, with a crescent moon, stars, birds and a sailboat out on the water. Along the pier stand a striped stall, a ticket hut, a helter-skelter tower, lamp posts and small groups of visitors, under strings of coloured bulbs that twinkle out of step with one another.

The wheel clears the stated trap exactly. Twelve gondolas ride a wheel that turns once every 36 seconds, each counter-rotating over the same 36 seconds so the two cancel, with a sway of four and a half degrees on top at three different periods so the cabins never swing in unison. The rim, the inner rings and the spokes all sit inside the rotating group, and 36 lamps chase round the rim with a bright head and a fading tail.

The carousel is the kind only one other model in this field has managed. Eight horses, two more than the brief asks for, travel round the ring: each sweeps across the ride, shrinks a little as it goes round the far side, narrows to nothing at either end and comes back mirrored, so it always faces the way it is moving. On the far side a horse passes behind the centre column, and on the near side in front of it. Each rides up and down its pole on a three-second cycle of its own. Above them, the canopy’s red gores sweep across the cone in the same direction and at the same rate as the horses beneath, one turn every twelve seconds, and the dividers on the board below the roof travel with them. The reflection is not a second drawing: the whole fairground is mirrored into the water as a live copy, so the wheel, the horses and the bulbs move in the water as they move above it. Every cycle in the file divides evenly into 36 seconds, so the scene loops without a seam.

Claude Opus 5.5’s scene, built in 62 minutes 41 seconds, set the standard this round is judged by. Its wheel carries twelve gondolas on a 24-second turn, each counter-rotating from its own starting angle with a four-degree sway, and its rim and bulbs turn with the spokes. Its carousel behaves like a solid object in perspective: six horses travel round an ellipse, each narrowing as it swings round the side and coming back mirrored, passing behind and in front of the centre column, while the roof’s stripes slide round the cone at exactly the horses’ speed. Side by side, the two builds are close relatives. Fable 5.2’s is the more richly drawn of the two, and that is the whole margin.

GPT-6 Astra’s has the finest horses anyone has drawn for this brief, with jewelled saddle blankets, bridles and flowing manes, in a pastel dusk with a lighthouse and a sailboat. Its Ferris wheel turns a full circle every 48 seconds with each gondola counter-rotating to stay level. Its carousel turns, and that is where it goes wrong: the six horses circle the column once every 12 seconds, but none of them ever turns to face the way it is going, so across the front of the ride they travel tail-first, and the canopy above them turns the opposite way.

Round 2 goes to Fable 5.2, which takes 5 points. Opus 5.5 takes 4 and Astra takes 3.

Round 3: The rocket launch

This brief leaves the method open. It names no library, asks for no 3D, and its only instruction on style is to make it look good. It wants a countdown, ignition, liftoff and a sustained ascent, with a mission-control HUD reporting the state of the flight.

AETHER-1 — Claude Fable 5.2, 47 minutes 1 second. From the terminal count to orbit in about two minutes.

HALCYON-3 — Claude Opus 5.5, 76 minutes 33 seconds. The whole flight takes about two minutes and ten seconds to play; stay for orbit.

GPT-6 Astra, 28 minutes.

Fable 5.2’s AETHER-1 flies the whole mission. A terminal count opens at T−13 with a launch director’s poll of six stations, each turning to GO in sequence. The engines light at T−3 under a sound-suppression deluge, the vehicle clears the tower at T+8 and starts its pitch program at T+14. It throttles down through maximum dynamic pressure at T+1:06, cuts its main engines at T+2:30, separates three seconds later and lights the second stage at T+2:40. The fairing halves fall away at T+3:20, it crosses 100 km three seconds after that, and at T+8:20 the second stage shuts down and the HUD confirms a 203 by 205 km orbit at 28,008 km/h, within a fraction of a percent of the speed a circular orbit at that height needs. Mission time runs up to eight times faster than real time through the long burns and drops back for staging and fairing separation, so the flight reaches orbit in about two minutes, coasts for a few seconds and starts again.

The HUD is thorough: dials for velocity and altitude, downrange distance, acceleration, Mach number, pitch, dynamic pressure, throttle, propellant gauges for both stages, the director’s poll, a running event log, a timeline of milestones and a caption for each event, with sound you can switch on. The flight path is a set of hand-placed keyframes and not a physics simulation, but its numbers agree with one another from the pad to orbit. In twelve frames sampled from the countdown to orbit, the vehicle is on screen in every one. The scene is handsome too: a dusk pad with a lattice tower and a lit vehicle, an exhaust cloud that swallows the pad, the curve of the Earth, and a sunrise breaking over the limb as the engine cuts off.

What it lacks is direction. The flight is filmed from one side-on tracking camera that pulls back steadily as the vehicle climbs, with a targeting frame round the rocket. It never cuts, and it never pushes in.

Claude Opus 5.5 flies the entire mission as well, in 76 minutes 33 seconds: a go/no-go poll, ignition at T−3, a throttle-down through peak aerodynamic pressure, main-engine cutoff at T+2:00, stage separation, second-stage ignition, fairing separation, and at T+7:53 a shutdown into a 203 by 207 km orbit at 28,024 km/h. The difference is in how it is shot. Opus 5.5 launches in twilight, cuts between angles, pushes in on the engines at ignition and pulls back as the vehicle climbs, and the vehicle is on screen at every moment we sampled. It plays like a broadcast with a director; Fable 5.2’s plays like the telemetry feed of the same flight.

GPT-6 Astra built a polished product and the shortest flight, in 28 minutes: a three-core vehicle rendered with real care, enormous editorial typography, the curvature of the Earth below, a zoom control and a playback-speed multiplier nobody asked for. Its ascent is modelled with some care too. And then it stops. Its sequence reads countdown, ignition, liftoff, max-Q, ascent, with no main-engine cutoff, no stage separation, no second stage and no orbit; the throttle stays at 100% until the animation runs out.

Round 3 goes to Opus 5.5, which takes 5 points. Fable 5.2 takes 4: the same complete mission, with less drama and far less camera work. Astra takes 2. Fable 5.2, too, drew everything by hand on a 2D canvas; the library the brief allows went unused again.

Quicker than Opus 5.5 on every brief

Build	Fable 5.2	Opus 5.5	Astra
Interchange	47m 17s	58m 49s	16m 15s
Fairground	41m 45s	62m 41s	26m 33s
Rocket launch	47m 01s	76m 33s	28m 00s
All three	136m 03s	198m 03s	70m 48s
List price per 1M tokens (in / out)	$8 / $40	$4 / $20	$10 / $50
Intelligence Index v5.0.1	65	59	57
Cost per index task	not yet published	$10.04	$6.24

Fable 5.2 needed about two and a quarter hours for the three builds, an hour less than Opus 5.5 and about twice as long as Astra. Like Opus 5.5, it spent most of that time thinking. On every brief its first response was 128,000 tokens of reasoning and nothing else, the most Claude Code allows in a single response, and it carried on reasoning after that: it began writing the launch page after 30 minutes, the fairground after 33 and the interchange after 42. Once it started, each file went down in pieces over five to seventeen minutes, and it finished each one by reading its own code back and correcting small things, the only kind of checking these rules allow.

That thinking is what the price buys. The three builds took about 763,000 output tokens between them, which comes to roughly $38 at Fable 5.2’s list price once the cached input is counted. The index figures in the table are from Artificial Analysis, each model at its maximum setting; it has not yet published what an index task costs on Fable 5.2.

The conditions matched where it mattered. Fable 5.2 received Opus 5.5’s prompts word for word apart from the save path, and like Opus 5.5 it ran in Claude Code at maximum effort with no browser and no way to run code. Astra ran in OpenAI’s Codex at the highest setting Codex offered at the time, under the same ban. None of Fable 5.2’s three builds was executed or tested before it was handed in.

The verdict: Fable 5.2 wins with 14 points, Opus 5.5 takes 12

	Interchange	Fairground	Launch	Total
Claude Fable 5.2	5	5	4	14
Claude Opus 5.5	3	4	5	12
GPT-6 Astra	4	3	2	9

Points were set round by round on the finished artefact alone, out of five. Opus 5.5’s and Astra’s builds are the files they were published with, scored again alongside Fable 5.2’s. The builds were scored at BitsMinds from a local side-by-side page of the untouched outputs.

Nothing left to catch

Most rounds in this series have been decided by a mistake: a line painted across an exit, a horse that never turns, a camera that loses its rocket. They are the faults a model would catch in seconds by looking at what it built, which is the one thing the format forbids. Fable 5.2 handed in three builds without one. The interchange opens its edge line at every ramp, the carousel turns as a single machine, and the launch runs from the count to a correct orbit with the vehicle in frame throughout. It got there by reasoning the picture through before writing it, and the files show it: the notes at the top of its interchange set out the lane positions, which lane feeds the ramps and the 24-second cycle before a single car is drawn.

Where it lost a point, it lost it on showmanship. Opus 5.5’s launch is directed and Fable 5.2’s is merely correct, and on a brief whose only style instruction is to make it look good, that counts. Astra remains the fastest of the three by a distance and the best at a striking first impression, with a mechanism missing in two rounds out of three. For Opus 5.5, the model that had won every test it entered here, two points is not a rout. It is still half the price per token. But on these three briefs the more expensive model was both better and faster, which is not how that trade usually goes.