AI video models
The models behind the apps. Most video apps resell a handful of models, so the model decides the quality and the app decides the price and workflow. Open each model to see which apps offer it.
Closed models (use through apps or APIs)
| Model | Maker | Released | Does | Audio | Apps we track |
|---|---|---|---|---|---|
| Gemini Omni Flash | May 2026 | Text to video, Image to video, Video editing | Yes | 2 | |
| Veo 3.1 | Not stated | Text to video, Image to video | Yes | 4 | |
| Wan 3.0 | Alibaba | August 24, 2026 | Text to video, Image to video, Document to video | Yes | 1 |
| Seedance 2.5 | ByteDance | July 31, 2026 | Text to video, Image to video, Region editing | Not stated | 2 |
| Kling 3.0 | Kuaishou | February 4, 2026 | Text to video, Image to video, Multi-shot | Yes | 4 |
| Runway Gen-4.5 | Runway | December 1, 2025 | Text to video, Image to video | Not stated | 2 |
Open-weight models (run on your own GPU)
Free to download, but each has its own license. Read it before commercial use.
| Model | Maker | Released | Does | Audio | Apps we track |
|---|---|---|---|---|---|
| MiniMax H3 (Hailuo 3) | MiniMax | July 31, 2026 | Text to video, Image to video | Yes | 2 |
| Wan 2.2 | Alibaba | 2025 | Text to video, Image to video | Not stated | - |
| LTX-2.5 | Lightricks | August 2026 | Text to video, Image to video | Yes | - |
| HunyuanVideo 1.5 | Tencent | Not stated | Text to video, Image to video | Not stated | - |
Which model is best?
Public leaderboards built on blind human votes (such as the Artificial Analysis video arena) reshuffle every few weeks, and snapshots disagree. In autumn 2026 Gemini Omni Flash, Wan 3.0, MiniMax H3 and Seedance 2.5 traded the top spots. Use the leaderboard for the current order, then pick by clip length, audio, price and license.
Last checked 2026-11-18. Each model page lists its sources.