If you picked "no idea," you're in the majority.

Most people can't name a Qwen image model.

FYO: Qwen-Image 3.0 is the real one. Canvas and Brush I just made up.

It’s weird because they're good at this. 2.0, back in Feb, topped the benchmarks, was open-weight, and free for anyone to run.

3.0, that just released the other day, now raises the bar. So I tried to break it.

Before that, some highlights:

The big open model everyone's posting about, Kimi K3, can build a whole game from a single sentence, and this pack shows you how, step by step.

Every build from the walkthrough, a physics roulette wheel, a GTA-style open world, an exploding interactive guitar, each with the exact prompt behind it.

One prompt in, a working website out, published to a link you can share the same day.

Qwen 3.0 is selling 3 things.

It shipped with no benchmarks, no open weights, and invite-only access.

A few of these are good.

Layout.

Both got all six panels right.

Qwen couldn't help itself with the line charts, a world map behind Countries, a ring gauge on Uptime, and stars under the rating.

GPT stuck to the prompt and gave me six plain cards.

The UI mockup.

English and CJK text.

So far, a real showing.

The languages.

English and the three CJK languages were clean.

But Qwen mangled Arabic and Hindi.

The Hindi has the wrong characters and the Arabic word for "future" comes out broken.

GPT Image 2 got both right.

So "twelve languages" holds right up until you pick one that isn't in the easy pile.

The number.

Qwen got the math right and then it printed the final figure as $1,157,63. At a glance that reads like 1.15 million.

The single number the entire graphic exists to show, and it flubbed the punctuation.

GPT printed "$1,157.63" clean, and showed its working on the way there.

The knowledge was right but the rendering wasn't.

Same story in the fine print.

On a tiny sources line, Qwen turned "Artificial Analysis" into mush while the bigger type stayed crisp, and GPT's fine print stayed readable.

So the 10px claim holds until the text gets small and long at the same time.

Qwen was never an image company. 2.0 was barely talked about because generating images isn't their game, so holding 3.0 up against GPT Image 2 is a little unfair.

Graded on that curve, it's a decent debut.

Am I blown away? No. If I'm honest.

Nothing here touches what the top models are doing, and there are plenty of misses.

But one thing tips it.

I made all of these on Qwen, for free. So, worth your time if you're on a budget.

Until next time,
Vaibhav 🤝🏻

If you read till here, you might find this interesting

#AD 1

Most finance teams still track SSP in spreadsheets — manually updated, disconnected from billing. On July 22, Tabs' product and marketing leads show what changes when SSP runs natively inside your billing platform: built from real billing data, connected to your Product Catalog, and generating audit-ready documentation automatically.

You'll leave with a clear view of how SSP and billing connect, what transaction price allocation looks like under ASC 606 in practice, and what it takes to close faster without reconciling SSP to contracts by hand.

July 22, 2026 · 1:30–2:00 PM EDT · Live + recording

Register now to save your spot, or read more about how Product Catalog powers billing automation.

Register for the Webinar

#AD 2

Skip the sales call. Walk through Gladly's interface yourself — the AI suggestions, the unified customer view, the full conversation thread. 15 minutes, no installation, no commitment.

See it yourself