Qwen4 Max vs GPT-6.1 Sol
Qwen4 Max took 18 min 15 s and $0.31 at list price, GPT-6.1 Sol took 12 min 10 s and $0.37 at list price. Qwen4 Max took 1.5× as long, and GPT-6.1 Sol cost 1.2× as much.
Qwen4 Max
Alibaba · $4.28/M out · 888 HN pts
GPT-6.1 Sol
OpenAI · $10.00/M out · 1,166 HN pts
Pelican
Generate an SVG of a pelican riding a bicycle
Watch
Draw a mechanical wristwatch that shows the current local time, with a sweeping seconds hand and a date window. Make it look expensive.
Landing
Build the landing page for Lumen Cycles, a Copenhagen workshop that builds custom cargo bikes to order. Make it look like a real brand you would trust with 4,000 euros.
Breakout
Make a Breakout game I can play right now with the keyboard or the mouse. Give it several levels, a score and some juice.
Logo
Design a logo for Night Heron, an independent bookshop that only opens after dark.
About
Write the opening of the About us page for Patchbay, a three-person studio in Rotterdam that repairs vintage synthesizers. About 150 words.
Patchbay is three people, one workshop, and a lot of schematics. We repair vintage synthesizers in Rotterdam West — the ones that have stopped sounding like themselves, and the ones that have stopped sounding at all.
Between us we've spent decades inside these machines: chasing dead voices in a Juno, replacing the battery that leaked before anyone noticed, rebuilding keybeds, recapping power supplies, tracking down the transistor nobody stocks anymore. We work from original service manuals where they exist and from careful measurement where they don't.
We're not trying to make old instruments look new. A worn panel is honest, and we'll leave it alone unless you ask. What we care about is that the thing plays — in tune, stable, reliable enough to take on tour or leave to a nephew.
Most of what comes through the door has been in a cupboard since 1994. We'd rather it were on a stage.
Patchbay is a three-person studio in Rotterdam, keeping vintage synthesizers in working hands. We repair the instruments that have picked up a crackle, lost a voice, drifted out of tune, or stopped making sound altogether.
We like the patient side of repair: tracing a fault, understanding an old circuit, and finding the right fix rather than the quickest workaround. Our aim is to make each instrument dependable without sanding away the character that made someone fall for it in the first place.
Some synths arrive with a clear problem. Others come with a story and a noise that only happens after twenty minutes. Either way, the work starts with listening—to the instrument and to the person who plays it.
We’re a small studio by choice. It keeps us close to the work, and to the musicians who trust us with something they want to keep making music with.
Specs and numbers
| Qwen4 Max | GPT-6.1 Sol | |
|---|---|---|
| Maker | Alibaba | OpenAI |
| Made in | China | US |
| Released | 16 Nov 2026 | 29 Sep 2026 |
| Price in, per M tokens | $1.43 | $2.00 |
| Price out, per M tokens | $4.28 | $10.00 |
| Context | 1,000,000 tokens | 1,050,000 tokens |
| Weights | closed | closed |
| Hacker News points | 888 | 1,166 |
| Code Arena | 1,750 · #9 of 160 | 1,744 · #10 of 160 |
| Text Arena | — | 1,483 · #27 of 422 |
| Time for the six prompts | 18 min 15 s | 12 min 10 s |
| List price of the six prompts | $0.31 | $0.37 |
Arena's ratings and ranks are external, from blind pairwise votes, and are not scores for our prompts. All the models here, ranked by Arena.
Qwen4 Max was cheaper: $0.053 less at list price, and less on 4 of the 6 prompts. GPT-6.1 Sol was faster: 6 min 5 s less in total, and faster on 5 of the 6 prompts.
In the recording, both Breakout games play.
Rebuild this screen experiment: Qwen4 Max's rebuild matched the screenshot at 62%, GPT-6.1 Sol's 97%. Tea Garden experiment: Qwen4 Max's game passed 5 of 5 checks, GPT-6.1 Sol's 5 of 5.





