Part of our AI Comparisons
Models·11 min read
By BitsMinds Hands-on test

GPT-6.2 Sol vs Claude Opus 5.5: The Mid-Tier Upset

GPT-6.2 Sol is OpenAI’s cheaper everyday model, so we gave it the three build briefs Claude Opus 5.5 has already answered. A motorway interchange, a seaside fairground and a cinematic rocket launch, one attempt each at maximum effort, with no way to see the result. All six builds run below, untouched, so you can judge them yourself.

GPT-6.2 SOL OPENAI OPUS 5.5 ANTHROPIC VS BITSMINDS.COM
Share:
BitsMinds LabHow this test was run2 models · 3 briefs · one attempt per model per brief · Maximum reasoning effort for both, at the top of each one's own dial: GPT-6.2 Sol in OpenAI Codex at max, Opus 5.5 in Claude Code at maxShow details
Models
GPT-6.2 Sol (OpenAI Codex) · Claude Opus 5.5 (Claude Code)
Attempts
One attempt per model per brief. No retries, no follow-up notes, no clean-up.
Reasoning effort
Maximum reasoning effort for both, at the top of each one's own dial: GPT-6.2 Sol in OpenAI Codex at max, Opus 5.5 in Claude Code at max.
Tools allowed
Each model ran inside its own maker's agentic coding environment and wrote its files to disk: GPT-6.2 Sol in Codex, Opus 5.5 in Claude Code. Both were told not to open a browser, preview the result or run code. Sol ran no commands at all; it only wrote its files.
Timing
Wall-clock time from prompt to finished file: GPT-6.2 Sol took 32m 56s, 23m 03s and 31m 43s on 18 November; Opus 5.5 took 58m 49s, 62m 41s and 76m 33s on 22 September.
Scoring
Points out of five, set round by round on the finished artefact. Opus 5.5's are the builds it was published with in September, scored again alongside Sol's. Final: GPT-6.2 Sol wins with 12 points, Opus 5.5 takes 11.
Judging
Scored by BitsMinds from a local side-by-side comparison page of the untouched outputs.
Published
November 18, 2026

Result

ModelProviderScore
GPT-6.2 SolOpenAI12 pts
Claude Opus 5.5Anthropic11 pts

GPT-6.2 Sol wins with 12 points, Opus 5.5 takes 11. Sol took both SVG rounds with builds in which nothing is broken, in less than half Opus 5.5's time; it lost the launch by two points, because its rocket climbs and never stages or reaches orbit.

The briefs — as described in the article; the exact prompt files were not published

  1. Round 1: Motorway interchange (SVG)

    A top-down view of a busy interchange as one self-contained SVG. Two roads crossing on different levels, connecting ramps, at least twelve vehicles following the curve of the road they are on, and correct layering so traffic on the lower road is hidden by the overpass.

    Result: GPT-6.2 Sol wins with 5 points, Opus 5.5 takes 3: both built four right-turn slip roads, but Sol's night scene opens its edge line into dashes at every ramp end, where Opus 5.5's solid edge line runs across all eight.

  2. Round 2: Seaside fairground (SVG)

    A seaside fairground at dusk, one SVG. A Ferris wheel with at least ten gondolas turning continuously, a spinning carousel with at least six horses rising and falling out of phase, twinkling bulbs, a reflection in the water. Each gondola must stay level with the horizon throughout the rotation.

    Result: GPT-6.2 Sol wins with 4 points, Opus 5.5 takes 3: both carousels turn as one machine with horses that face their travel and pass behind the centre column; Sol's eight-horse scene is the better drawn.

  3. Round 3: Rocket launch (web page)

    A self-contained web page that animates a rocket launching — countdown, ignition, liftoff and ascent — with a mission-control HUD. No library named, no 3D requested, one aesthetic instruction: make it look good.

    Result: Opus 5.5 wins with 5 points, Sol takes 3: Opus 5.5 flies the whole mission to orbit with a director's camera; Sol's handsome page ends at the ascent, with no cutoff, no staging and no orbit.

Read with care

  • Single run per model — a re-run could land differently.
  • Opus 5.5's builds are the files it was published with in September, scored again here rather than re-run.
  • Scored on the finished artefact only. Wall-clock time is reported but carries no points.

Part of BitsMinds Lab, our series of original hands-on tests.

OpenAI released GPT-6.2 Sol on 17 November alongside GPT-6.2 Astra. Sol is the cheaper of the two, the model OpenAI builds for coding and everyday agent work, and it kept GPT-6.1 Sol’s price of $2 per million input tokens and $10 per million output. On the Artificial Analysis Intelligence Index it scores 57, two points behind Claude Opus 5.5, and an index task costs about $1.01 on it against $10.04 on Opus 5.5. This series tests a model a different way: the same build briefs as its rival, one attempt, and the result published untouched.

Its opponent brings the builds it was published with. Claude Opus 5.5, Anthropic’s everyday flagship and the model that led this series from September until Claude Fable 5.2 arrived in November, built them for its 22 September round. 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.

GPT-6.2 Sol wins with 12 points and Opus 5.5 takes 11. Sol took both SVG rounds with builds in which we found nothing broken, and lost the launch by two points, because its rocket climbs and never does anything else. It is the first time an OpenAI model has finished ahead of Opus 5.5 in this series, and Sol needed less than half Opus 5.5’s time to do it.

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.

Animated motorway interchange built in one attempt by GPT-6.2 Sol
GPT-6.2 Sol, 32 minutes 56 seconds. Forty-eight vehicles at night, a slip road in each quadrant, and lane lines that open where every ramp leaves or joins.
Animated motorway interchange built in one attempt by Claude Opus 5.5
Claude Opus 5.5, 58 minutes 49 seconds. Thirty-two vehicles, four ramps, and a turn signal on every car that takes one.

GPT-6.2 Sol set its interchange at night, in dark woodland beside a lake, with a solar farm in one corner and a depot with a car park in another. A six-lane highway runs diagonally across the frame and a second one crosses it underneath, curving gently on either side of the bridge so the junction does not read as a grid. Twenty-six lamp posts throw pools of light along the shoulders, and every vehicle carries its own pair of headlight beams: thirty cars, thirteen estates and five lorries, forty-eight in all.

Its layout is the same kind Opus 5.5 chose: four slip roads, one in each quadrant, each carrying traffic from the outer lane of one highway to the outer lane of the other, so the junction offers right turns and no left ones. Three cars use each ramp. The traffic is organised by lane: every vehicle belongs to one of fourteen streams, each stream runs on its own cycle of 10, 12, 15 or 20 seconds, and the scene repeats exactly once a minute. Sampled every fifth of a second across a full minute, no vehicle jumps anywhere in the frame, and each ramp car follows the curve of its slip road from one highway to the other. The cars on the lower road are drawn before the bridge and its cast shadow, so they pass under the deck and disappear, and the paths for the ramps and the upper highway start and end off the canvas, so nothing appears out of thin air.

The road markings are where it beats Opus 5.5. Where each slip road leaves or joins a highway, the solid edge line gives way to a short dashed section across the mouth of the ramp, the way a real road marks a slip road, at every ramp end. That is exactly the detail Opus 5.5 got wrong. We could not find a fault in Sol’s build.

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.

Round 1 goes to GPT-6.2 Sol, which takes 5 points: a striking picture in which everything we checked behaves as it should, including traffic passing under the bridge. 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.

Animated seaside fairground built in one attempt by GPT-6.2 Sol
GPT-6.2 Sol, 23 minutes 3 seconds. Eight horses travel round the carousel facing the way they go, and the harbour mirrors both rides as they turn.
Animated seaside fairground built in one attempt by Claude Opus 5.5
Claude Opus 5.5, 62 minutes 41 seconds. The carousel turns as one machine, and its horses turn to face the way they are going.

Sol’s fair stands on a wooden pier under a violet sky, with a large apricot sun sinking behind the carousel, headlands across the harbour and a lighthouse in the distance. A ticket kiosk and a striped stall close off the two ends of the pier, and four strings of bulbs hang twinkling across the scene.

The wheel clears the stated trap. Twelve cabins hang from a truss that turns once every 48 seconds, and each cabin counter-rotates over the same 48 seconds, so the two cancel at every suspension point, with a sway of two and a half degrees either way on top. The rim and its lights turn with the spokes, and only the hub badge stays still.

The carousel is where most models in this series have gone wrong, and Sol’s is right. Eight horses ride round the deck once every 12 seconds, each on a path that runs across the front of the ride and back behind the centre column. As a horse reaches the side of the ride it narrows to nothing, and on the far side it comes back mirrored and slightly smaller, so it always faces the way it is moving. Each horse is drawn twice, once behind the column and once in front of it, and the two copies swap at the side of the ride, where the horse has no width and the change cannot be seen. Above them the canopy’s painted panels sweep round on the same 12-second cycle, and each horse rides its pole up and down on a four-second rhythm of its own. The reflection is the live scene mirrored into the water, so the wheel and the horses move in the harbour as they do on the pier.

Claude Opus 5.5’s scene, built in 62 minutes 41 seconds, is the one this round has long been 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. The two carousels work the same way. Sol’s is the better-looking scene.

Round 2 goes to GPT-6.2 Sol, which takes 4 points: no faults, and the richer picture. Opus 5.5 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.

Aurora 01 — GPT-6.2 Sol, 31 minutes 43 seconds.

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.

Sol’s Aurora 01 is a clean, quiet piece of design. A white launcher stands beside its tower on a dark headland, with the sun low on the sea behind it. The mission clock is set in large type over the picture, with a status line beneath it, and a strip along the bottom tracks altitude, speed, g-load and downrange distance, each with a small running chart. A four-step timeline marks countdown, ignition, liftoff and ascent, and there are pause and replay buttons. The engines light at T−0, the hold-downs release about two seconds later in a billow of exhaust, and the camera follows the vehicle up through the dusk and into black sky above the curve of the Earth, pulling back as it climbs and keeping the rocket in frame throughout.

Then nothing else happens. The last of the four steps is “Ascent nominal”, and the flight stays there: the engine never cuts off, no stage separates, and there is no orbit. Three minutes in, the HUD reads 157 km at 4,003 km/h, roughly a seventh of the speed an orbit at that height needs, and the rocket is still climbing. There is no sound. The GPT-6 Sol and GPT-6.1 Sol builds of this brief also flew no cutoff and reached no orbit.

Claude Opus 5.5 flies the entire mission, 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. It launches in twilight, cuts between camera 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.

Round 3 goes to Opus 5.5, which takes 5 points for the full mission and the drama it is shot with. Sol takes 3: a handsome page and a single act, with no staging, no camera work and nothing after liftoff but the climb. Sol drew everything by hand on a 2D canvas; the library the brief allows went unused again.

Less than half the time

BuildGPT-6.2 SolOpus 5.5
Interchange32m 56s58m 49s
Fairground23m 03s62m 41s
Rocket launch31m 43s76m 33s
All three87m 42s198m 03s
List price per 1M tokens (in / out)$2 / $10$4 / $20
Intelligence Index v5.0.15759
Cost per index task$1.01$10.04

Sol needed just under an hour and a half for the three builds, against more than three hours for Opus 5.5, and it was quicker on every brief, most of all on the fairground, where it finished in well under half the time. It wrote each file in one or two passes. By Codex’s own count the three builds used about 149,000 output tokens, which comes to a little over $2 at Sol’s list price once the cached input is included.

The index figures in the table are from Artificial Analysis, each model at its maximum setting. On that measure Sol is two points behind Opus 5.5 at about a tenth of the cost per task, and on these three briefs it came out a point ahead.

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

The verdict: GPT-6.2 Sol wins with 12 points, Opus 5.5 takes 11

InterchangeFairgroundLaunchTotal
GPT-6.2 Sol54312
Claude Opus 5.533511

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

Two clean pictures and one missing act

Sol won this on the two briefs that are decided by details. Its interchange gets the road markings right at every slip road, which Opus 5.5’s does not, and its carousel solves the same problem Opus 5.5’s does, with eight horses that face their travel and a canopy that turns with them, in a scene drawn with more care. Neither SVG has a mistake we could find, and that has been rare on these briefs for any model, at any price.

The launch shows where the gap still is. Opus 5.5 read “sustained ascent” as the start of a mission and flew it to orbit, with a director’s camera. Sol read it as the last word of the brief and stopped there, in a page that looks finished and plays one act. One point is not much of a margin, and it rests on the score of a single round. But the model that won it costs half as much per token and a tenth as much per index task, and it got there in less than half the time.

More on Claude

Evergreen coverage we keep current — start here.

Want AI news before everyone else?

The morning's most important AI stories, straight to your inbox. No fluff.

Related Articles