Key Takeaway
The Artificial Analysis Text-to-Video Leaderboard (No Audio) now tracks 73 models. HappyHorse-1.0 leads at ELO 1,382 with 14,507 blind comparison samples — 108 points ahead of the second-place Seedance 2.0 (1,274). Data accessed April 14, 2026.
Top 20 Rankings — Text-to-Video (No Audio)
Rankings are based on ELO scores calculated from blind user votes on the Artificial Analysis Video Arena. Users are shown two videos generated from the same prompt and choose which one they prefer, without knowing which model produced which video.
| # | Creator | Model | ELO | Samples | API Price |
|---|---|---|---|---|---|
| 1 | Alibaba-ATH | HappyHorse-1.0 | 1,382 | 14,507 | Coming soon |
| 2 | ByteDance Seed | Seedance 2.0 720p | 1,274 | 8,513 | No API |
| 3 | KlingAI | Kling 3.0 1080p (Pro) | 1,242 | 5,698 | $13.44/min |
| 4 | Skywork AI | SkyReels V4 | 1,242 | 6,440 | $7.20/min |
| 5 | KlingAI | Kling 3.0 Omni 1080p | 1,228 | 5,144 | $13.44/min |
| 6 | xAI | grok-imagine-video | 1,227 | 6,169 | $4.20/min |
| 7 | Vidu | Vidu Q3 Pro | 1,223 | 6,543 | $9.60/min |
| 8 | Runway | Gen-4.5 | 1,222 | 8,908 | No API |
| 9 | PixVerse | PixVerse V5.6 | 1,221 | 5,768 | $9.00/min |
| 10 | KlingAI | Kling 3.0 Omni 720p | 1,218 | 5,159 | $10.08/min |
| 11 | Veo 3 | 1,217 | 3,813 | $12.00/min | |
| 12 | KlingAI | Kling 3.0 720p | 1,214 | 5,470 | $10.08/min |
| 13 | KlingAI | Kling 2.5 Turbo 1080p | 1,210 | 4,584 | $4.20/min |
| 14 | PixVerse | PixVerse V6 | 1,210 | 16,749 | $5.40/min |
| 15 | Veo 3.1 Fast | 1,209 | 4,942 | $6.00/min | |
| 16 | Veo 3.1 | 1,209 | 5,445 | $12.00/min | |
| 17 | Luma Labs | Ray 3 | 1,203 | 4,819 | $13.20/min |
| 18 | KlingAI | Kling O1 Pro | 1,203 | 5,815 | $10.08/min |
| 19 | KlingAI | Kling 2.6 Pro | 1,196 | 5,201 | $4.20/min |
| 20 | OpenAI | Sora 2 Pro | 1,194 | 5,114 | $30.00/min |
Source: Artificial Analysis Text-to-Video Leaderboard (No Audio), accessed April 14, 2026. Full leaderboard includes 73 models.
Key Observations
The HappyHorse Gap Is Enormous
The 108-point gap between HappyHorse (1,382) and Seedance 2.0 (1,274) is the largest gap between #1 and #2 in the leaderboard's history. For context, the gap between #2 and #10 is only 56 points. HappyHorse is not just leading — it is in a category of its own based on blind user preference.
KlingAI Dominates by Volume
KlingAI has 10 separate model variants in the top 25, more than any other creator. This reflects their strategy of offering multiple tiers (Standard, Pro, Omni, Turbo) across different resolutions and generations. Their highest-ranked model, Kling 3.0 1080p Pro, sits at #3 with ELO 1,242.
Google Veo Holds Steady
Google has four Veo variants in the top 20 (Veo 3, Veo 3.1, Veo 3.1 Fast, and the older Veo 3 Fast Preview at #26). Despite being released in mid-2025, Veo 3 remains competitive at ELO 1,217 — a testament to its quality, though its $12/min pricing makes it one of the more expensive options.
OpenAI Sora Falls Behind
Sora 2 Pro sits at #20 with ELO 1,194 — nearly 190 points behind HappyHorse. At $30/min, it is also the most expensive model in the top 20 by a wide margin. The original Sora is at #49 with ELO 1,045. OpenAI's video generation efforts have not kept pace with the competition.
New Entrants to Watch
Three models were added to the leaderboard in the last month: HappyHorse-1.0, Dreamina Seedance 2.0 720p, and PixVerse V6. SkyReels V4 from Skywork AI is also a notable new entrant, tied for #3-4 with Kling 3.0 at ELO 1,242.
Price vs. Quality
For developers evaluating cost-effectiveness, the leaderboard reveals some interesting value propositions:
| Model | ELO | API Price | Value Assessment |
|---|---|---|---|
| xAI grok-imagine-video | 1,227 | $4.20/min | Best value in top 10 |
| SkyReels V4 | 1,242 | $7.20/min | Strong value for top-5 quality |
| PixVerse V6 | 1,210 | $5.40/min | Good mid-range option |
| Sora 2 Pro | 1,194 | $30.00/min | Poor value — 7x more expensive than grok |
| LTX-2 Pro (open weights) | 1,129 | $3.60/min | Best open-weights option, self-hostable |
Methodology Note
The Artificial Analysis Video Arena uses an ELO rating system based on blind pairwise comparisons. Users see two videos generated from the same prompt and vote for their preferred result without knowing which model produced which video. The system has collected over 400,000 total comparisons across all models. ELO scores fluctuate as new votes come in — the numbers in this article reflect a snapshot taken on April 14, 2026.
Sources
- Artificial Analysis — Text to Video Leaderboard (accessed April 14, 2026) artificialanalysis.ai