The headline features that hold up
Native audio is the one that matters. Sora 2 generates lipsync dialogue, footfalls, ambient room tone and light musical stings in the same pass as the visual. It isn't perfect — dialogue over four seconds still drifts — but a five-second shot with a spoken line, correctly lipsynced, is now a single API call.
Physics: better, not solved
Sora 2 handles gravity, cloth and liquid convincingly in short shots. It still fails on complex collisions (two people high-fiving remains a lottery). The upgrade is enough that 'unusable' becomes 'usually fine' — a huge economic swing.
Where Veo 3.1 still wins
For rigorous shot direction — 'dolly-in from three-quarter, rack focus to the mug, lens flare at 4s' — Veo 3.1 is still tighter. Sora 2 interprets, Veo 3.1 executes. Ad agencies notice; they're keeping Veo in the pipeline.
Pricing reality
At launch, Sora 2 was priced to signal premium: roughly $1.20 for a 10-second clip on aggregators, more on the direct API. Two months in, that dropped to under $0.60. Expect another halving before year-end as capacity comes online.
What to actually do this week
Two experiments: replace one voiceover-plus-B-roll social post with a native-audio Sora 2 shot, and try a lipsync dialogue moment you would previously have avoided. Both will change how you brief future work.
Generate with Sora 2
Put this into practice in the studio — under a minute to your first result.
Generate with Sora 2 →