HappyHorse-1.0 just took the AI video crown.
On 2026-04-07 a brand-new model called HappyHorse-1.0 appeared on the Artificial Analysis leaderboards and immediately took the top spot on both text-to-video and image-to-video — and has held #1 ever since. Current scores: T2V Elo 1,389 (+115 over Seedance 2.0) and I2V Elo 1,416 (+58 over Seedance 2.0).
The developer is now officially listed as Alibaba-ATH on the leaderboard itself. Multiple Chinese media outlets confirmed the team: Alibaba's ATH (AI Technology & HappyHorse) unit, led by Zhang Di — the ex-Kuaishou VP who built Kling 1.0 and 2.0 — who returned to Alibaba in November 2025. The team officially launched on 2026-04-10. Data verified 2026-04-10 from artificialanalysis.ai.
🐎 Unofficial / informational site. Not affiliated with the developer of HappyHorse-1.0.
The numbers, sourced from one place
Every figure on this page is read from a single config file with a verified-on date — when the leaderboard updates, this whole site catches up with one commit. Data verified 2026-04-10 against artificialanalysis.ai.
1,389 Text-to-Video Elo (+115 over Seedance 2.0, the runner-up)
Text-to-Video Elo (+115 over Seedance 2.0, the runner-up)
1,416 Image-to-Video Elo (+58 over Seedance 2.0, the runner-up)
Image-to-Video Elo (+58 over Seedance 2.0, the runner-up)
#1 / #1 Rank on both Artificial Analysis boards · 2026-04-10
Rank on both Artificial Analysis boards · 2026-04-10
Top 5 — Text → Video
Source: artificialanalysis.ai · Data verified 2026-04-10 · Higher Elo is better.
1. HappyHorse-1.0
Elo 1,389 · Developer: Alibaba-ATH · Leads by +115
2. Dreamina Seedance 2.0 720p
Elo 1,274 · Developer: ByteDance Seed
3. SkyReels V4
Elo 1,244 · Developer: Skywork AI
4. Kling 3.0 1080p (Pro)
Elo 1,243 · Developer: KlingAI (Kuaishou)
5. Kling 3.0 Omni 1080p (Pro)
Elo 1,230 · Developer: KlingAI (Kuaishou)
Top 5 — Image → Video
Source: artificialanalysis.ai · Data verified 2026-04-10 · Higher Elo is better.
1. HappyHorse-1.0
Elo 1,416 · Developer: Alibaba-ATH · Leads by +58
2. Dreamina Seedance 2.0 720p
Elo 1,358 · Developer: ByteDance Seed
3. grok-imagine-video
Elo 1,333 · Developer: xAI
4. PixVerse V6
Elo 1,311 · Developer: PixVerse
5. Kling 3.0 Omni 1080p (Pro)
Elo 1,300 · Developer: KlingAI (Kuaishou)
What we know — with citations
Facts about HappyHorse-1.0 that are publicly verifiable. Every line links back to its source.
Developer confirmed: Alibaba-ATH (listed on leaderboard)
The Artificial Analysis leaderboard now lists the developer as 'Alibaba-ATH'. Multiple Chinese media also confirmed the team: Alibaba's ATH unit, originating from the Taotian Group (淘天集团) Future Life Lab — the entity behind Taobao and Tmall.
Lead: Zhang Di (张迪)
The project is led by Zhang Di, former VP of Technology at Kuaishou and the architect of Kling 1.0 / 2.0. He left Bilibili and returned to Alibaba in November 2025. Source: Sina Finance, Zhidongxi, multiple Chinese outlets.
Team spun out as independent entity
The team has since been spun out of Taotian Group into an independent company structure. Source: multiple Chinese media reports, April 2026.
Official launch: 2026-04-10
The team announced an official public launch on 2026-04-10. Source: 163.com exclusive report, Sina Finance.
T2V leaderboard rank
Ranked #1 on the Artificial Analysis text-to-video board with Elo 1,389, leading by +115 points over Seedance 2.0. Source: artificialanalysis.ai, verified 2026-04-10.
I2V leaderboard rank
Ranked #1 on the Artificial Analysis image-to-video board with Elo 1,416, leading by +58 points over Seedance 2.0. Source: artificialanalysis.ai, verified 2026-04-10.
First appeared on leaderboard: 2026-04-07
First appeared on both Artificial Analysis boards on 2026-04-07. Source: artificialanalysis.ai 'added in the last month' section.
API status
API listed as 'Coming soon' on artificialanalysis.ai. No public endpoint exists yet. We will update the moment it does.
What we don't know — and won't pretend to
The developer identity is now confirmed, but many technical details remain unverified by an official source. Sites that confidently quote specs are repeating community estimates, not official documentation.
Architecture (widely reported, not officially confirmed)
Community sources widely report 15B parameters, a 40-layer Transformer processing text/image/video/audio tokens in one sequence, and 8-step DMD-2 distillation. However, the developer has published no technical paper or official spec sheet. We won't treat community reports as official facts.
Output resolution & speed (widely reported)
1080p video in ~38 seconds on a single H100 is widely cited, but this figure has no official source. Treat as unconfirmed until the team publishes a spec.
Native audio & lip-sync (widely reported)
Multiple sources claim the model generates synchronized audio and video in one pass and supports multilingual lip-sync. Not yet confirmed by an official technical report.
Training data
Composition, source, and licensing of training data are not disclosed.
License & weights
Widely reported as open-source with all weights to be released. As of 2026-04-10 no weights are publicly available and both GitHub and Model Hub links show 'coming soon'. We will update when released.
Pricing
No tier structure, per-second cost, or free tier has been disclosed.
API endpoint
URL, auth scheme, rate limits — none of it exists publicly yet.
Paper or technical report
No paper, blog post, or official technical report has been published as of 2026-04-10.
Get one email when the API ships
Drop your email and we'll send exactly one message — the day HappyHorse-1.0's official API becomes publicly available. No newsletter. No marketing. No reselling. Read our short privacy policy for the details.
Frequently asked questions
Short answers about HappyHorse-1.0 and this site.
One email. The day the API drops.
No drip campaign, no upsell, no resale of your address. Just the announcement.