Veo 3.1 vs Sora 3 in real client work
We ran 40 shots through both models on three real client briefs. Here's the honest cost-vs-quality breakdown — including the moment we switched mid-project and what we'd do again.
Three weeks ago a global beverage brand asked us for a 6-shot cinematic sequence — perfume-tier polish, 1080p, delivered in seven business days. Half the brief screamed Veo. Half screamed Sora. So we ran both. Same prompts. Same references. Same constraints.
The setup
40 shots total: 18 product hero shots, 14 motion-heavy reveal shots, 8 "atmosphere" cuts. We used identical prompt grammar across both models, the same negative-prompt set, and the same seed-iteration loop.
Where Veo wins
Reflections, metals, transparent fluids. Anything where light physics has to feel "expensive" — Veo nailed it 8/10 generations. Sora 3 got there too but needed 3× the iterations.
- Reflective product hero shots: Veo by 2-3 generations
- Liquid pour shots: Veo by a wide margin
- Cinematic depth-of-field control: tied
Where Sora 3 wins
Cloth physics, hair movement, human micro-expression. The model just understands soft-body simulation better. For anything with a person in frame, Sora was our default.
The mid-project switch
On day 4 we realized the "atmosphere" shots — those soft, slow, almost-still environmental moments — were eating our Veo budget without delivering the painterly feel the brand wanted. We switched those 8 shots to Sora 3 entirely. Cost dropped 60%. Quality went up.
The lesson: model selection isn't per-project. It's per-shot. Build a quick test matrix on day one and let the numbers decide.
The verdict (with numbers)
Veo 3.1 averaged $0.62/successful shot at 1080p Fast tier. Sora 3 averaged $0.48/successful shot. Both within budget for client work that bills $200-400/shot.
Both models are now permanent in our stack — we just stopped treating them as competitors. They're tools with different strengths. The interesting question isn't which one is "better." It's which one for THIS shot.
