An AI video model appeared from nowhere on April 7, 2026. No announcement. No press release. No corporate logo. Just a model called "HappyHorse-1.0" quietly climbing the Artificial Analysis Video Arena leaderboard until it sat at #1 in both text-to-video and image-to-video generation worldwide.
Three days later, Alibaba claimed it.
The Hong Kong-listed tech giant's shares rose 2.12% on the day of the reveal, according to CNBC. But the real story isn't just a benchmark win — it's the collapse of Western competition at the exact moment Chinese labs are finding their stride. OpenAI discontinued Sora earlier this year. ByteDance's Seedance 2.0 hit a copyright wall with Hollywood studios and stalled. Into that vacuum walked HappyHorse, armed with a radically different technical architecture and a stealth launch strategy that let the model win on pure merit before anyone knew who built it.
This is how the AI video wars are being fought in 2026. And Alibaba just scored a decisive victory.
What Happened: The Stealth Launch That Shook the Industry
HappyHorse-1.0 first appeared on the Artificial Analysis benchmark platform around April 7 under no company affiliation. It climbed fast, briefly disappeared from the leaderboard with no official explanation, then returned with Alibaba's confirmation on April 10.
The blind test scores tell the story. In text-to-video with no audio, HappyHorse-1.0 scored 1333 Elo — beating the previous leader by 60 points. In image-to-video with no audio, it hit 1392 Elo, 37 points ahead of its nearest rival. Across all four benchmark categories, it ranks either #1 or #2:
- Text-to-Video (no audio): 1333 Elo — #1
- Image-to-Video (no audio): 1392 Elo — #1
- Text-to-Video (with audio): 1205 Elo — #2
- Image-to-Video (with audio): 1161 Elo — #2
The model isn't the unanimous winner across every dimension — it ranks #2 in audio-video categories. But in silent video generation, which reflects raw visual quality and motion realism, the results are dominant.
The disappearance from the leaderboard before Alibaba's reveal wasn't a technical glitch. From the timeline, this appears to have been a controlled pause before a planned announcement — a deliberate stealth play: drop the model anonymously, let it win on merit in blind human tests, then claim it once the scores were locked in.
The strategy worked. By the time Alibaba put its name on HappyHorse, the model had already proven itself against every competitor in the world.
Why the Timing Matters: The Collapse of Western Competition
HappyHorse-1.0 is landing at the exact moment Western competition has collapsed, which makes this benchmark win far more significant than a typical technical achievement.
OpenAI discontinued Sora earlier in 2026 to focus on AGI and enterprise tools. The official reasoning was that video generation was too compute-heavy for the return it delivered. Whether that's the full story or not, the result is the same: the company that kicked off the AI video craze with Sora's viral demo in February 2024 has exited the consumer video generation business entirely.
ByteDance's Seedance 2.0 was supposed to fill the gap. Instead, Seedance hit a copyright wall with Hollywood studios and paused its global rollout. The Chinese tech giant's video generation tool, which had been gaining traction, is now stuck in regulatory and legal limbo.
What makes this striking is that two of the strongest non-Chinese challengers have now either exited or stalled at roughly the same time, leaving Alibaba positioned to dominate by default. The realistic top-tier options right now are HappyHorse, Google's Veo, and Kuaishou's Kling AI. From the blind test data, HappyHorse has a clear benchmark lead over Kling in the categories that matter most for visual fidelity.
This isn't just a technical win for Alibaba. It's a market opportunity that opened up because the competition faltered at the exact moment Alibaba was ready to strike.
How It Works: The Technical Breakthrough
HappyHorse-1.0 uses a 40-layer single-stream Self-Attention Transformer that processes text, video, and audio as one unified token sequence — skipping the Cross-Attention structures most competing models rely on.
From a technical standpoint, this is the most interesting part of the story. Most multimodal video models use separate processing pathways for text, video, and audio, with Cross-Attention bridging them. HappyHorse drops that entirely. All three token types go into one sequence and attend to each other directly, with the middle 32 of the 40 layers sharing parameters across all modalities.
The design pays off in speed. The model requires only 8 denoising steps and no Classifier-Free Guidance to reach its output quality, where most current video models use 20 to 50 steps. On an H100 GPU, HappyHorse generates a 5-second 256p clip in roughly 2 seconds, and a 1080p version in 38 seconds.
The model runs at an estimated 10B to 30B parameters (the official site claims 15B), supports six languages natively, and is specifically optimized for human-centric scenarios: lip-sync accuracy, facial performance, and realistic body movement.
There's been speculation that HappyHorse is a rebrand of Alibaba's Wan 2.7 model. From the technical specs, the architectures don't match. Wan 2.7 focuses on long-text rendering and "thinking modes." HappyHorse uses the single-stream Transformer described above. These appear to be two parallel research bets running inside Alibaba at the same time — and HappyHorse is the one that just paid off.
The Open Source Question: What "Fully Open" Actually Means
The official landing pages claim the model is "fully open-source" with base, distilled, and upscaling versions released. In practice, the GitHub and HuggingFace links both show "Coming Soon" as of April 10.
The gap between the marketing claim and actual code availability is the most important caveat right now. Anyone delaying live projects waiting for the release should reconsider — the timeline is unclear, and the "open-source" claim may turn out to mean something narrower than it sounds, similar to corporate model releases that restrict commercial use.
When access does arrive, three things are worth checking before committing:
- Licensing terms: Is it MIT/Apache 2.0, or a restricted commercial license?
- Content moderation rules: What does Alibaba restrict under Chinese regulatory requirements?
- Training data transparency: What footage was used to achieve the lip-sync accuracy it's being praised for?
Chinese AI video tools have been pricing at roughly 4 cents per second generated, which undercuts Western alternatives by a wide margin. If HappyHorse delivers on its open-source promise with permissive licensing, it could become the default foundation model for AI video generation in 2026.
🔥 Our Hot Take: This Is How You Win the AI Wars
Alibaba just executed the perfect AI product launch — and every Western lab should be taking notes.
The stealth strategy was brilliant. By releasing HappyHorse anonymously, Alibaba removed every possible source of bias from the benchmark results. There was no "Alibaba effect" inflating scores. No nationalist cheering. No pre-existing brand loyalty. Just a model competing purely on quality against every competitor in the world — and winning.
Compare this to how Western labs launch products. The hype cycles. The countdown timers. The carefully staged demos that may or may not reflect real performance. The inevitable gap between marketing promise and shipping reality. Alibaba skipped all of it. They let the work speak for itself, then claimed the credit once the verdict was in.
The technical architecture is equally smart. The unified transformer approach isn't just faster — it's conceptually cleaner. While competitors are duct-taping together separate pathways for text, video, and audio, Alibaba built something that treats all modalities as the same kind of data from the ground up. That's the kind of architectural bet that pays dividends for years.
But here's the part that should worry Western AI labs: this is happening while their own video generation efforts are collapsing. OpenAI abandoned Sora because video wasn't profitable enough. ByteDance's Seedance is stuck in copyright hell. The field is wide open, and a Chinese lab just claimed the top spot with a model that appears to be genuinely superior.
The open-source question is the only thing holding this back from being an absolute slam dunk. If Alibaba follows through with truly permissive licensing, HappyHorse could become the Linux of AI video — the default foundation that everyone builds on. If they restrict it or delay release, they'll squander the momentum they've built.
Either way, the message is clear: Chinese AI labs aren't just catching up. In video generation, they've just taken the lead — and they did it by being smarter about how they compete, not just how they build.
The AI video wars aren't over. But Alibaba just fired the shot that everyone else will be responding to for the next year.