OpenAI's Sora went from the most talked-about video model in the industry to a company playing catch-up in about ten months. OpenAI unveiled it in February 2024 with demo reels that made every competitor look unfinished, then largely sat on it while the rest of the field kept shipping. By the time a limited version, Sora Turbo, reached ChatGPT Plus and Pro subscribers in December 2024, Runway, Kling, and Luma already had paying customers and real production use cases.
According to The Decoder, the AI video generation market has since recovered from that stumble. A sector that briefly looked like it might stall out on one company's slow rollout is instead expanding, with several vendors now shipping tools people actually pay for rather than one lab sitting on a demo reel.
That recovery matters more than the Sora news cycle itself. It says something about how quickly this category can route around a single player's mistakes — and it's a preview of how fast the ground can shift under any AI builder who treats a launch delay as low-risk.
A ten-month gap changed the leaderboard
Sora's problem was never the output quality in isolation — it was the gap between the demo and the product. OpenAI showed the model off in February 2024 and then kept it behind a red-teaming wall for most of the year, fueling a narrative that OpenAI announces things it can't yet ship at scale. Competitors used the runway:
- Runway shipped Gen-3 Alpha to paying users in mid-2024, well before Sora had any public access.
- Kling AI, from China's Kuaishou, launched to strong early reception and iterated fast on resolution and motion consistency.
- Luma AI's Dream Machine and Pika's successive releases gave creators cheap, fast options with none of the waitlist friction Sora carried.
When Sora Turbo finally arrived, it landed with a 20-second clip cap, visible watermarking, and quality that read as solid rather than category-defining — a hard comparison against demo footage that had set expectations a year earlier.
Who actually filled the gap
Google was the biggest beneficiary of the delay. Veo 2 launched in December 2024, and Veo 3 followed in mid-2025 with native synchronized audio generation — dialogue, sound effects, and ambient noise produced alongside the video rather than bolted on afterward. That one feature reset expectations for what "finished" output looks like, and it pushed every other vendor to treat audio as part of the core product rather than a post-production step.
OpenAI did eventually push back with Sora 2 and a companion social app in late 2025, but that relaunch brought its own problems: backlash over deepfakes of real and deceased people and generated clips using copyrighted characters, which forced OpenAI into reactive opt-in and rights-holder opt-out changes after launch rather than before it. The market didn't wait around for OpenAI to get either version right — it kept moving.
What this means if you're building on these tools
For teams shipping AI video features, the practical lesson isn't "which model wins" — it's that no single vendor is a safe default anymore. A few things worth building around now:
- Don't hardcode one vendor. Runway, Kling, Luma, Veo, and Sora have different strengths, pricing, and access models (API vs. app-only); an abstraction layer that can swap providers protects you from the next rollout stumble or price change.
- Plan for audio-native generation. Once one major vendor ships synced audio by default, pipelines built around silent clips plus a separate TTS or foley step start to look like extra work, not a feature.
- Budget for provenance and watermarking. C2PA-style content credentials and visible watermarks are becoming standard across vendors, not an OpenAI-only quirk — plan for them in your output pipeline rather than stripping them as an afterthought.
- Treat likeness and rights handling as a launch requirement, not a patch. Sora 2's post-launch scramble on deepfakes and copyrighted characters is the clearest recent example of what happens when consent and rights-holder controls ship after the product instead of with it.
AiiN's takeaway
The interesting part of this story isn't that Sora stumbled — big launches slip. It's that the AI video category didn't wait for OpenAI to fix it. Runway, Kling, Luma, and Google all had working products in front of paying users during the exact window when Sora was still a waitlist. That's a useful data point for anyone assuming a strong demo buys a durable lead: in fast-moving AI categories, shipping late to a market that has other credible options is close to not shipping at all. In our estimation, the vendor that eventually wins the video category will be decided less by who had the best February 2024 demo and more by who ships audio, rights controls, and API access together, on schedule, without a second false start.