@Alibaba_Qwen just dropped qwen-image-3.0 – the third gen of their image model. the whole pitch is going from "good-looking" to "useful" (their words):
• rich content (dense multi-element layouts)
• authentic details (10px legible text)
• target use cases: newspaper pdfs, storyboards, exam papers, product layouts
so we built the test straight from their own claims:
1. retail promo flyer (walmart-style circular) – logo, offer, price cards
2. one-page fine-dining menu with dish photos
3. dark data-dashboard infographic – depth scale + bar chart + donut + hero stat
4. broadsheet newspaper column – "spain crowned world champions, beat argentina 1-0"
5. high-fashion magazine cover with pull-quote and credits
per-image winner (gpt time / qwen time):
• flyer – gpt (59s / 2m02s)
• menu – gpt (1m05s / 3m11s)
• infographic – gpt, close (1m08s / 2m49s)
• newspaper – gpt (1m09s / 3m52s)
• fashion – toss-up (1m02s / 1m15s)
observations:
1. gpt's flyer looks like a real circular: fuller produce crate, yellow price tags fused onto each product shot. qwen's is cleaner but reads generic
2. ironic one: qwen's entire pitch is text rendering, yet it misspelled the menu – "thryme," "rosted fennel" – while gpt kept every word clean, accents and all ("crème fraîche")
3. both nailed the ocean data (-10,935 m, the 40/500/730/3,800 bar chart, 5% mapped). gpt edged it with sharper creature art
4. the newspaper is the tell. gpt wrote a full, legible, plausible article – morata scoring in the 63rd off a pedri pass, messi's shot tipped over by unai simón, spain's 2nd title after 2010. qwen's body copy is pure gibberish, and it put the players in striped kits that read as argentina, not spain
5. where qwen genuinely closes the gap: the fashion cover and the dashboard. type scale, palette control and composition are legit – legible, on-brief, well-built
verdict: gpt image 2 still wins on fine detail and text coherence. but qwen image 3.0 is a genuine leap: coherent dense layouts, mostly accurate text, strong design instincts. good work by the qwen team – the gap to the frontier just got a lot smaller
🎨 Meet Qwen-Image-3.0 — the third generation of our foundational image generation model.
— Qwen (@Alibaba_Qwen) July 22, 2026
If 1.0 was about "Precision," and 2.0 added "Variety, Completeness, Beauty & Authenticity," then 3.0 comes down to a single word: Real (实).
Three dimensions of "Real":
📰 Rich Content —… pic.twitter.com/YkaBy47Qkm
Nick Trenkler