Best AI Video Generators in 2026 (Honest Comparison)

June 2, 2026 · Updated September 30, 2026 · 9 min read

As of September 30, 2026, the video models that win blind comparisons are Google's Gemini Omni Flash, Alibaba's Wan 3.0 and MiniMax H3, with ByteDance's Seedance 2.0 right behind them. Which one is best for you depends on clip length, resolution, price per second and where you can actually use it — and one former favorite is gone: OpenAI removed Sora from its API on September 24, 2026.

The short list (checked September 30, 2026)

Rank and Elo come from Artificial Analysis's Text to Video Arena with audio, where a recruited panel picks between two clips made from the same prompt without knowing which model made them. “Longest clip” is the longest single generation the maker documents. These are third-party and vendor figures — we haven't benchmarked the models ourselves.

#3 on the leaderboard is a fal-hosted variant of MiniMax H3. Not ranked yet: Seedance 2.5 (released July 31, 2026) and Kling 4.0 (announced September 28, 2026). Specs are from each maker's documentation; rankings move every few weeks.

What changed since the spring

• Sora is gone. OpenAI announced on March 24, 2026 that it was discontinuing Sora. The app closed on April 26, and the sora-2 and sora-2-pro API models were removed on September 24 with no replacement named. That settles Sora 2 vs Veo 3.1.
• 30-second single takes arrived. Seedance 2.5 (July 31) and Wan 3.0 (August) generate up to 30 seconds in one pass. Kuaishou announced Kling 4.0 with the same limit on September 28; a lighter 4.0 Flash (720p, up to 20 seconds) opened to Ultra annual subscribers first, and the full model is due in October.
• Google's next model isn't called Veo 4. At I/O on May 19, 2026, Google launched Gemini Omni Flash, which now tops the arena. Veo 3.1 is still sold — and still labeled a preview model on the Gemini API.
• Sound is standard. Every model in the table generates audio with the picture, and the arena now ranks clips with audio on their own board.

The new leaders: Gemini Omni Flash, Wan 3.0, MiniMax H3

Gemini Omni Flash takes text, images, video and audio as input and lets you edit a clip by describing the change. Version 1.1 (late August) generates 3–10 seconds per call. It's rolling out in the Gemini app, Google Flow and YouTube Shorts, and on the Gemini API it costs about $0.10 per second of 720p video. The catches: short clips and strict filters — Artificial Analysis notes it refused some of its editing prompts, most often ones involving people.

Wan 3.0 (Alibaba, August 2026) makes up to 30 seconds at up to 1080p with native audio in one pass, from text, images, video or audio. It ranks first on the arena's no-audio board (1,335). Its weights aren't public: you reach it through Alibaba Cloud Model Studio or platforms that offer it, such as Pika.

MiniMax H3 (late July 2026) produces 4–15 seconds at up to 2K and 24 fps with stereo sound, in six aspect ratios from 21:9 to 9:16. It's the highest-ranked model with public weights (on Hugging Face, under MiniMax's community license); hosted access is through MiniMax's API and its Hailuo app. None of these three is available on AI Growth Kit.

Seedance vs Veo vs Kling: the old rule of thumb, rechecked

The standard advice used to be: Seedance for volume, Veo for the cinematic hero shot, Kling for stylized motion. Blind votes no longer support Veo as the quality pick — Seedance 2.0 ranks 5th (1,210), Kling 3.0 Pro 12th (1,095) and Veo 3.1 14th (1,088). What still separates them is length, resolution and price.

Seedance 2.0 (ByteDance)

Released in February 2026, it generates 4–15 seconds with sound from a prompt plus up to 12 reference files (at most 9 images, 3 videos and 3 audio clips). BytePlus lists it at $0.15 per second at 720p, $0.37 at 1080p and $0.78 at native 4K; the Fast and Mini variants cost $0.12 and $0.08 at 720p, their maximum. Seedance 2.5 (July 31, 2026) doubles the length to 30 seconds — see what Seedance 2.0 is and what 2.5 changed.

Veo 3.1 (Google)

Launched October 15, 2025 and updated in January. Clips are 4, 6 or 8 seconds in 16:9 or 9:16 at 24 fps, and 1080p and 4K are only available at 8 seconds. Its strengths now are 4K output and a long extension chain: 7 seconds at a time, up to 148 seconds in total (at 720p). Google's Gemini API sold it until October 22, 2026, when Google shut down the Veo 3.1 Standard, Fast and Lite preview endpoints; Veo 3.1 and 3.1 Fast remain on Google's Gemini Enterprise Agent Platform. Runway's API prices both per second, as of November 1, 2026:

• Veo 3.1: $0.40/s with audio, $0.20/s without (40 / 20 credits).
• Veo 3.1 Fast: $0.15/s with audio, $0.10/s without (15 / 10 credits).

With audio the standard tier costs about 2.7× Fast, so the old habit — draft on Fast, render the final once — still works inside Veo itself.

Kling 3.0, with 4.0 on the way (Kuaishou)

Kling 3.0 (February 2026) generates 3–15 seconds at 720p or 1080p, with optional native audio and a multi-shot mode where you set each shot and its length. In Kling's own app, 1080p costs 12 credits per second with audio and 8 without. Kling 4.0, announced September 28, raises the limit to 30 seconds and adds up to 10 keyframes.

Bottom line: Seedance 2.0 is the best-ranked of the three for short clips. Veo 3.1 is the cheaper route to 4K ($0.30–$0.60/s against Seedance's $0.78), has the longest extension chain, and its Lite tier is the cheapest of the three at 720p. Kling 3.0 fits when you want several planned shots inside one 15-second generation.

HappyHorse and Grok Imagine

HappyHorse 1.0 appeared anonymously on Artificial Analysis in April 2026 and was then confirmed as Alibaba's. It makes up to 15 seconds of 1080p video with multiple shots, through Alibaba Cloud Model Studio and the Qwen app. It ranks 9th with audio but 4th without (1,286); version 1.1 (June) ranks 8th with audio.

Grok Imagine comes from SpaceXAI, the name xAI has used since July 2026, and has had a public API since January 28. Version 1.5 makes up to 15 seconds and can animate an image, follow reference images, or edit and extend a clip. Listed from $0.05 per second, it's among the cheapest options; it ranks 23rd with audio and 8th without.

Runway vs Pika: both are model marketplaces now

The old comparison — Runway's production suite versus Pika's quick social clips — matters less now that both resell other companies' models. Compare the plans:

• Runway. Its newest in-house generation model is still Gen-4.5 (December 1, 2025), at 12 credits per second. Runway documents where it fails: effects that happen before their causes, objects that appear or vanish, and actions that succeed when they shouldn't. Standard is $15/month ($12 billed yearly) for 625 credits — about 52 seconds of Gen-4.5. Pro is $35 ($28) for 2,250 credits and adds Seedance 2.5; Max is $95 ($76) for 9,500. Monthly credits don't roll over on Standard or Pro.
• Pika. Pika 2.5 sits next to hosted Seedance 2.5, MiniMax H3, Wan 3.0 and Google Omni Flash. Starter is $10/month ($8 yearly) for 900 credits, Creator $35 ($28) for 3,150, Fancy from $95. The free plan includes no monthly credits and no commercial license; every paid plan, Starter included, has one. A 5-second 720p clip costs 122 credits on Seedance 2.5 or 33 on Wan 3.0; top-ups are 60 credits per $1.

Runway suits you if you'll use its editor and Gen-4.5; Pika is the cheaper entry point.

Which one should you use?

• Best blind-vote results: Gemini Omni Flash for clips up to 10 seconds, Wan 3.0 if you need up to 30.
• One continuous 30-second shot: Wan 3.0 or Seedance 2.5 today; Kling 4.0 once it fully ships.
• Generate on your own GPUs: MiniMax H3.
• 4K delivery: Veo 3.1 ($0.60/s, or $0.30/s on Fast) or Seedance 2.0's 4K tier ($0.78/s).
• Lowest listed prices: Seedance 2.0 Mini ($0.04/s at 480p), Veo 3.1 Lite ($0.05/s at 720p) and Grok Imagine (from $0.05/s).
• Models inside a full creative app: Runway (from $15/month) or Pika (from $10/month).
• Several models, no API keys, no subscription: our Video Generator, within the limits below.

Where AI Growth Kit fits — and where it doesn't

Our Video Generator runs five models from one form: Seedance 2.0 (the default), Kling 3.0, Veo 3.1, Grok Video (SpaceXAI's Grok Imagine) and Happy Horse 1.0. The menu names each version but not its tier, so don't assume Veo 3.1 Standard rather than Fast. The form offers only what the chosen model renders: 4–15 seconds on Seedance 2.0, 3–15 on Kling 3.0 and Happy Horse 1.0, 4, 6 or 8 on Veo 3.1 and 6 or 10 on Grok Video; 16:9 and 9:16 on every model, 1:1 on all but Veo, 4:3 and 3:4 on Seedance and Happy Horse; 720p everywhere and 1080p on all but Grok.

Every generation costs a flat 10 credits per second, whatever the model or resolution — 50 credits for a 5-second clip, 150 for 15 seconds. Credits come in one-time packs that never expire ($9 for 200, $29 for 750, $59 for 1,800) or monthly plans ($19 for 600, $49 for 2,000), which works out to roughly $0.25–$0.45 per second of video. The 20 free credits you get at sign-up don't cover a clip: the shortest, 3 seconds, costs 30. Details are on the pricing page.

What we don't offer: Gemini Omni Flash, Wan 3.0, MiniMax H3, Seedance 2.5 or Kling 4.0; image-to-video or reference uploads (the form is text-only); audio settings; 4K; or clips longer than 15 seconds. To lengthen a clip, Video Extend continues it with Seedance 2.0 at 10 credits per second.

Is it cheaper? Not always. If you need one model and can work with an API, going direct usually costs less at 720p — BytePlus lists Seedance 2.0 at $0.15 per second, and Veo 3.1 Fast is $0.10. The flat rate pays off when you switch models per shot without separate accounts, or already use our translation, subtitle or upscaling tools on the same balance. For prompts that carry over between models, see how to write AI video prompts.