How do the AI video models compare?
| Model | Maker | Max clip | Max resolution | Audio | Open weights | API from |
|---|---|---|---|---|---|---|
| Veo 3.1 | 8 s | 4K | Yes | No | $0.03/s | |
| Gemini Omni Flash | 15 s | 4K (upscaled) | Yes | No | $0.023/s | |
| Seedance 2.5 | ByteDance | 30 s | 1080p | Yes | No | $0.103/s |
| Kling 4.0 | Kuaishou (Kling AI) | 30 s | 1080p (4K announced) | Not stated | No | $0.094/s |
| MiniMax H3 | MiniMax (Hailuo) | 15 s | 2K | Yes | Yes | $0.025/s |
| Runway Gen-4.5 | Runway | 10 s | 720p | No | No | $0.12/s |
| Luma Ray3.2 | Luma AI | 10 s | 1080p HDR | No | No | Credits (no $ rate) |
| Pika H3 Flex | Pika | 15 s | 4K (upsampled) | Yes | Adapters | $0.033/s |
| Wan 3.0 | Alibaba | 30 s | 1080p | Not stated | Wan 2.2 only | $0.05/s |
| Sora 2 (shut down) | OpenAI | 20 s | 1080p (Pro) | Yes | No | Was $0.10/s |
From each developer's own docs and pricing, checked November 10, 2026: Veo 3.1, Gemini Omni Flash, Seedance 2.5, Kling 4.0, MiniMax H3, Runway Gen-4.5, Luma Ray3.2, Pika H3 Flex, Wan 3.0, Sora 2 (sources on each page). "API from" is the lowest per-second list price published for any tier or resolution, so it isn't like for like. Kling 4.0's price is Runway's API (Kling's own API lists only 3.0); Pika's is computed from its Starter plan credits.
Which AI video model should you use for which job?
| Job | Models | Why |
|---|---|---|
| Longest single shot | Seedance 2.5, Kling 4.0, Wan 3.0 | 30 seconds in one generation (Kling 4.0 also extends to 2 minutes). |
| Dialogue and sound effects in the clip | Veo 3.1, Gemini Omni Flash, Seedance 2.5, MiniMax H3 | Audio is generated with the video; Seedance 2.5 and MiniMax H3 each list 11 dialogue languages. |
| Editing footage you already have | Seedance 2.5, Gemini Omni Flash, Kling 4.0, Luma Ray3.2 | Video input with edit or restyle modes; Runway's Aleph 2.0 (on the Gen-4.5 page) does the same at $0.28 a second. |
| 4K delivery | Veo 3.1, Gemini Omni Flash, Pika H3 Flex | Veo 3.1 (full model) outputs 4K; Omni Flash and Pika H3 Flex upscale to it. Seedance 2.0 and Kling 3.0 also do 4K; their newer versions don't yet. |
| Colour grading and compositing | Luma Ray3.2 | Native 16-bit HDR and EXR export. |
| Cheapest drafts | Gemini Omni Flash, MiniMax H3, Veo 3.1 | From $0.023/s (Gemini Omni Flash, 360p, Gemini API), $0.025/s (MiniMax H3, H3-Lite, 480p), $0.03/s (Veo 3.1, Lite, 720p, video only). |
| Running it on your own GPUs | MiniMax H3, Wan 3.0, Pika H3 Flex | MiniMax H3 (community license), Wan 2.2 (Apache 2.0) and Pika H3 Flex adapters publish weights. Wan 3.0 itself is hosted only. |
| A launch or explainer video of your own software | None of them | Generated clips can't show your real interface. Use a motion graphics or launch video tool instead: see AI motion graphics tools and AI explainer video generators. |
What does 30 seconds of AI video cost?
Computed from each model's published per-second rate at 720p (or the nearest tier), for one take. Real projects need several takes per shot.
| Model | Tier | Per second | Generations for 30 s | 30 seconds |
|---|---|---|---|---|
| Veo 3.1 | Veo 3.1 Fast, 720p, with audio | $0.10 | 4 | $3.00 |
| Gemini Omni Flash | 720p | $0.07 | 2 | $2.10 |
| Seedance 2.5 | 720p | $0.231 | 1 | $6.93 |
| Kling 4.0 | Runway API, 720p (to Nov 30) | $0.126 | 1 | $3.77 |
| MiniMax H3 | MiniMax-H3, 768p | $0.08 | 2 | $2.40 |
| Runway Gen-4.5 | Gen-4.5 | $0.12 | 3 | $3.60 |
| Pika H3 Flex | H3 Flex 768p, Starter monthly | $0.033 | 2 | $1.00 |
| Wan 3.0 | 720p | $0.10 | 1 | $3.00 |
Luma prices its API in credits without a dollar rate, so it's left out. Rates and sources are on each model's page.
What changed in AI video models in 2026?
- Sora 2 is gone. OpenAI shut down the Sora 2 models and the Videos API on September 24, 2026. (OpenAI API: Sora 2 model page; details)
- Veo moved to Google Cloud; Omni Flash took over the Gemini API. Veo 3.1's Gemini API previews shut down on October 22, 2026, and Google points to Gemini Omni 1.2 Flash (released November 5). Veo 3.1 stays GA on the Gemini Enterprise Agent Platform. (Gemini API: deprecations; details)
- 30-second clips arrived. Seedance 2.5 (4 to 30 s), Kling 4.0 (3 to 30 s) and Wan 3.0 (up to 30 s) all make half-minute shots in one pass; Seedance 2.0 and Kling 3.0 stopped at 15 s. (BytePlus ModelArk: Seedance 2.5 tutorial; details)
- Open weights got serious. MiniMax H3 (33B parameters) is on Hugging Face, and Pika released H3 Flex adapters on top of it. (Hugging Face: MiniMaxAI/MiniMax-H3; details)
Where can you use several models at once?
Runway's API lists 29 video models from several developers, priced at $0.01 a credit (models, pricing). Pika's app hosts Seedance 2.5 and 2.0, Kling 4.0, Wan 3.0, MiniMax H3 and Gemini Omni 1.1 Flash next to its own models (Pika pricing). Among general video tools, invideo, HeyGen, Fliki, Pictory, Renderforest and Powtoon name specific models on their pricing pages; each model page lists which. Compare those tools on alternatives and head-to-head comparisons.
All AI video model pages
- Veo 3.1: Google, up to 8 s, 4K
- Gemini Omni Flash: Google, up to 15 s, 4K (upscaled)
- Seedance 2.5: ByteDance, up to 30 s, 1080p
- Kling 4.0: Kuaishou (Kling AI), up to 30 s, 1080p (4K announced)
- MiniMax H3: MiniMax (Hailuo), up to 15 s, 2K
- Runway Gen-4.5: Runway, up to 10 s, 720p
- Luma Ray3.2: Luma AI, up to 10 s, 1080p HDR
- Pika H3 Flex: Pika, up to 15 s, 4K (upsampled)
- Wan 3.0: Alibaba, up to 30 s, 1080p
- Sora 2: OpenAI, shut down September 24, 2026
Choosing a finished tool rather than a model? See the best AI video generators by use case, the free AI video generators and their exact limits, the best AI video editing apps, or the step-by-step guide to making an AI video.
