GPT-6.1 Sol vs GPT-6 Astra and GPT-6 Sol: A Dead Heat

OpenAI released GPT-6.1 Sol on 29 September with a big claim: that it nearly matches GPT-6 Astra on coding, computer use and professional work at one fifth of Astra’s price per token, $2 and $10 per million tokens against $10 and $50. On the Artificial Analysis Intelligence Index it already comes close, 52 against Astra’s 53. So we put the claim to the three briefs this series knows best, and set it against the two models it sits between: Astra, the flagship above it, and GPT-6 Sol, the model it replaces at the same price.

The briefs are a motorway interchange seen from above and a seaside fairground at dusk, each as one animated SVG, and a cinematic rocket launch as one web page. GPT-6.1 Sol ran on 29 September in OpenAI’s Codex, all three briefs at once, on the same prompts GPT-6 Sol and GPT-5.6 Sol were given, word for word apart from the path each file was saved to. It ran at max reasoning effort, with no browser, no preview and no way to run code, so it never saw what it had made. Its opponents bring the builds they were published with: Astra’s from its 6 September head-to-head with Claude Fable 5.1, and GPT-6 Sol’s from its 22 September round against Astra and GPT-6 Luna.

GPT-6.1 Sol and Astra finished level on 9 points each. GPT-6 Sol took 4.

Round 1: The interchange

The brief asks for a busy interchange 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 of its ends, and whether traffic keeps to the right side after it merges.

GPT-6.1 Sol took 15 minutes 22 seconds to build a complete cloverleaf, seen from directly above. An east–west highway crosses on a bridge deck over a north–south highway, and eight ramps link them: four loops and four long sweeping turns. The corners are landscaped with retention pools and gardens inside the loops, low buildings with solar-panel roofs, footpaths and trees kept clear of the roads. Forty-four vehicles, cars, SUVs, lorries and buses, drive on the right, consistently, past direction arrows painted on the asphalt and striped gore markings where the ramps meet the roads. Every route starts and ends outside the frame, so no car appears out of nothing, and the layering holds: traffic on the lower road passes beneath the deck, and loop cars heading for the upper road stay above it. The fault is in how the ramps join. None has a lane of its own where it meets a highway. Each ramp route runs inside the highway’s own lane for part of its length, on the same line as the through traffic, with its own separate timing: the north-west loop, for one, travels 816 pixels along the westbound outer lane before it turns off. So ramp cars and through cars drive straight through each other. Over four minutes we counted 62 moments when two visible vehicles sat almost on top of one another, about 15 a minute and the longest lasting 3.5 seconds, and every one of them involved a ramp or loop car.

GPT-6 Astra’s interchange is still the busiest this series has seen. Forty-eight vehicles, an articulated lorry and a bus among them, run along four straight through lanes and four curving ramps, and every one throws a headlight cone onto the road in front of it. Direction arrows are painted on the asphalt, there is street lighting down the median, road signs, two ponds and a river, and the bridge is a proper structure with a shadowed deck, parapets and bearings at its four corners. Every path begins off-screen, so ramp traffic arrives along the main road before it curves away, and the occlusion holds even halfway across: a lorry comes in from the west with its cab already under the parapet and its trailer still in the open. The fault is a painted one. The white shoulder line runs straight across the exit, so the ramp never visibly separates from the road it leaves.

GPT-6 Sol built its interchange in 4 minutes 45 seconds. Twenty-four vehicles, sedans, vans and trucks with headlamps and red tail lights, drive on a ground-level east–west highway, across a north–south overpass and along four single-lane ramps, through planted verges, clumps of trees and drainage lines. The ramps are raised, each casts a shadow, and the east–west traffic passes beneath them. It is their ends that fail. At the lower road each ramp starts as a rounded stub in the middle of the carriageway instead of peeling off a lane, and at the other end the ramp traffic is drawn underneath the overpass deck, so cars heading up to the bridge disappear beneath it instead of joining its lanes.

Round 1 goes to Astra, which takes 4 points; GPT-6.1 Sol takes 3 and GPT-6 Sol takes 1. GPT-6.1 Sol has none of its predecessor’s mess and the better-looking plan of the two Sols, but its ramps never separate from the highway and its cars keep running into each other.

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 carousel with at least six horses rising and falling out of phase under a spinning canopy, twinkling bulbs, lights chasing round the rim of the wheel, a low sun or moon, and the whole scene reflected in the water. The gondolas are the trap the brief states. The carousel is the one it does not.

GPT-6.1 Sol spent 21 minutes 27 seconds on a pier it titled “The last light at Luna Pier”: a ticket booth, a candy-floss stall with a bunch of balloons, a second kiosk, families strolling along the boards, a lighthouse on a point, a sailboat, a low sun and strings of bulbs slung across the sky. Its twelve-cabin wheel stands on an A-frame and turns once every 40 seconds while each cabin counter-rotates through a full circle over the same 40 seconds and sways on its own, so the stated trap is cleared. The carousel is where it moves past both of the others. Its striped canopy turns every 12 seconds, and six horses on travelling poles circle the column on the same clock, rising and falling out of phase and passing behind the column and in front of it. Canopy and horses go round together, in the same direction. The horses still never turn to face the way they are going: each points right the whole way round, so across the front of the ride they travel tail-first. And the reflection works, a compressed, rippling mirror image of the whole pier in the water.

GPT-6 Astra’s fairground is a pastel dusk with a lighthouse and a sailboat on the water, and its carousel horses wear jewelled saddle blankets and bridles under flowing manes. They are the best-drawn horses this brief has produced. The Ferris wheel passes the stated trap cleanly: it turns once every 48 seconds while each gondola counter-rotates through a full circle over the same 48 seconds, with a gentle sway on top, so the cabins stay level. The carousel turns, and that is where it goes wrong. The six horses circle the column once every 12 seconds, rising and falling on their poles, but none of them ever turns to face the way it is going: every horse points right for the whole revolution, so across the front of the ride they travel tail-first. The canopy above them turns the opposite way, its stripes sweeping left to right across the front while the horses beneath them move right to left.

GPT-6 Sol’s scene, built in 6 minutes 18 seconds, is a pier at sunset under a band of clouds, with a low sun, a headland, a ticket kiosk, a “SWEETS” taffy booth, lamp posts, bunting and two long festoons of bulbs that twinkle independently. Its twelve-cabin wheel turns once every 48 seconds while each cabin counter-rotates and sways, so the stated trap is cleared, and two rings of chasing lights run round the rim. Its carousel turns too: a striped canopy rotates every twelve seconds and eight horses ride an elliptical track at the same speed, rising and falling at staggered offsets. What they never do is turn, so on the far side of the ride every horse travels backwards. It wrote a reflection, then clipped it with a rectangle placed in the flipped coordinates of the reflected scene, which cuts every reflected ride away. The water shows ripples and nothing else.

Round 2 goes to GPT-6.1 Sol, which takes 3 points; Astra takes 2 and GPT-6 Sol takes 1. Astra still has the finer drawing, but GPT-6.1 Sol’s ride turns as one machine and its reflection is really there.

Round 3: The rocket launch

The only open brief. It names no library, asks for no 3D, and its one 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.

ASTERIA 01 — GPT-6.1 Sol, 30 minutes 2 seconds.

GPT-6 Astra, 28 minutes 0 seconds.

AURORA 06 — GPT-6 Sol, 6 minutes 36 seconds. Each starts on its own countdown.

GPT-6.1 Sol’s ASTERIA 01 lifts off from “Cape Aurora, Pad 07” in 30 minutes 2 seconds of work, the longest it spent on any brief. It is laid out like a product page: large editorial type for the current phase, a four-step automatic sequence that marks each stage GO, LIVE or WAIT, a mission-control voice loop that logs each call as it happens, and a telemetry panel with altitude, velocity, thrust, axial load and dynamic pressure over a velocity graph, plus pause, replay and sound controls. A twelve-second count leads into ignition at T−3.2. The loop logs tower clear at T+9, the pitch programme at T+27 and max Q at T+54, where the throttle dips by 12% and comes back at T+63. At T+78 the two side boosters separate and fall away in full view, and the camera keeps the vehicle in frame through all of it. Then the flight stops developing. After booster separation nothing else happens: no main-engine cutoff, no second stage, no orbit. The core keeps burning at 100% under a constant acceleration that never ends, so five minutes in the rocket is 478 km up at 3.4 km/s and the HUD still reads “powered ascent”. It does not have Astra’s zoom control.

GPT-6 Astra built a polished product: a three-core vehicle rendered with care, enormous editorial type, the curve of the Earth below, a flight-sequence timeline, a zoom control and a playback-speed multiplier that nobody asked for. The ascent is modelled rather than decorative. Its 38.5 km at 1,003 metres per second by T+1:39 is close to a real climb, and the g-load rises from 2.13 to 3.29 as propellant burns off, which is the right behaviour rather than a busy-looking number. Then its flight stops developing too. The sequence runs countdown, ignition, liftoff, max Q and ascent, with no main-engine cutoff, no stage separation, no second-stage burn and no orbit. The throttle stays at 100% and the vehicle climbs until the animation runs out.

GPT-6 Sol’s AURORA 06 follows much the same plan in a quarter of the time. A three-core rocket stands beside its service tower against dark mountains, under a headline that reads “The sky awaits”, and a ten-second count leads into three seconds of ignition, with flame and smoke building under the pad. The HUD reports altitude, velocity and downrange distance on a four-step phase track. Once the vehicle clears the tower the camera holds it at a fixed height while the ground falls away, but the shot is too tight, and the rocket’s nose sits cut off at the top of the frame for the whole climb. The flight never goes past “ascent”: the engines stay at 100% and nothing separates.

None of the three pulled in a library, though the brief allows any from a public CDN. Like every model that has taken this brief since September, GPT-6.1 Sol drew its launch by hand on a 2D canvas.

Round 3 is a draw between GPT-6.1 Sol and Astra, which take 3 points each; GPT-6 Sol takes 2. GPT-6.1 Sol separates its boosters, which Astra never does, and Astra answers with the zoom control and the richer flight. Both still end at “ascent”.

Astra’s time, at a fraction of Astra’s price

Build	GPT-6.1 Sol	GPT-6 Astra	GPT-6 Sol
Interchange	15m 22s	16m 15s	4m 45s
Fairground	21m 27s	26m 33s	6m 18s
Rocket launch	30m 02s	28m 00s	6m 36s
All three	66m 51s	70m 48s	17m 39s
API list price, per million tokens in / out	$2 / $10	$10 / $50	$2 / $10
Intelligence Index (v4.3.2, max)	52	53	48
Cost per index task	$0.72	$3.26	$1.05

GPT-6.1 Sol is not a faster GPT-6 Sol. It took three to four and a half times as long as its predecessor on every brief, and across the three it needed 66 minutes 51 seconds, within four minutes of Astra’s 70 minutes 48 seconds. It also wrote more: 92,938 output tokens by Codex’s own count, against GPT-6 Sol’s 55,902, and files two to three times the size. Priced at the list rate from the token counts Codex reports, its three builds come to about $1.33, against about 76 cents for GPT-6 Sol. The runs went through a ChatGPT subscription, so no one was billed those amounts; they show how the price tag plays out on the same work.

The saving over Astra is where the price matters. On Artificial Analysis’s Intelligence Index, as our leaderboard shows, GPT-6.1 Sol scores 52 at max effort for $0.72 per task, against Astra’s 53 for $3.26: nearly the same score for about a fifth of the cost. On the index it is also cheaper per task than GPT-6 Sol, $0.72 against $1.05 at the same per-token price. On our three briefs it went the other way, with more tokens and a bigger bill than its predecessor.

The conditions were the same where it mattered: the same prompts word for word, max effort for both Sols, no browser and no code execution for all three. Astra’s builds date from 6 September, when it ran at xhigh, then the top of Codex’s dial; Codex has since added max, the setting both Sols ran at. Wall-clock time also depends on the server as well as the model, and the three models ran on different days. GPT-6.1 Sol’s Codex logs show nothing but file writes: one file per brief, written to the exact path it was given.

The verdict: GPT-6.1 Sol and GPT-6 Astra draw with 9 points each, GPT-6 Sol takes 4

	Interchange	Fairground	Launch	Total
GPT-6.1 Sol	3	3	3	9
GPT-6 Astra	4	2	3	9
GPT-6 Sol	1	1	2	4

Points were set round by round on the finished artefact alone: 4, 3 and 1 on the interchange, 3, 2 and 1 on the fairground, and 3, 3 and 2 on the launch. Astra’s and GPT-6 Sol’s builds are the files they were published with, scored again beside GPT-6.1 Sol’s, which is why their points differ from earlier rounds. The builds were scored at BitsMinds from a local side-by-side page of the untouched outputs, with no blind scoring and no automated grading. GPT-6 Sol’s builds are weighed against the benchmarks in our GPT-6 Sol review; its step up from GPT-5.6 Sol was measured on these same briefs, and the 29 September round on them, Claude Sonnet 5.5 against Opus 5.5, Astra and GPT-6 Sol, shows where the Claude models stand.

The claim holds, and the clock tells you why

On these three briefs, OpenAI’s claim stands up. GPT-6.1 Sol finished level with its flagship on points, beat it on the fairground, drew with it on the launch, and lost only the interchange, and it more than doubled the score of the model it replaces at the same price. Every one of its faults is still the kind a single look would catch: ramp cars running through the traffic they merge into, horses riding tail-first, a flight that never gets past “ascent”. They are fewer and smaller than GPT-6 Sol’s.

What it did not do is get there cheaply in time. The old Sol was the quick one, done with all three briefs in under 18 minutes; the new one worked for as long as Astra did. That looks like the trade OpenAI made: Sol now takes as long as Astra and scores like Astra, and what it kept from the old Sol is the per-token price.