AI Video Model Leaderboard 2026: Top 12 Models Ranked Today

This video model ai leaderboard ranks 12 AI video models by public arena scores from two sources, Arena and Artificial Analysis, both read on 8 November 2026. Then it adds the columns the boards skip: clip length, native sound, resolution and price per second.

We compared the published boards and vendor figures. We didn’t run these models ourselves.

For specs, see our AI video model specs comparison, and for the basics, what video AI models are.

How this video model ai leaderboard is ranked

The rank is the model’s score on Arena’s text-to-video leaderboard, which shows “last updated Nov 6, 2026”, 891,470 votes and 50 models. We read it on 8 November 2026.

Each row also carries the same model’s score on Arena’s image-to-video board (also updated Nov 6, 2026) and on both Artificial Analysis text-to-video (AA-Video-T2V v2.1) and image-to-video (AA-Video-I2V v2.1). Artificial Analysis shows no date on those pages, so we give the day we read them, 8 November 2026.

Artificial Analysis AA-Video-T2V v2.1, with Wan 3.0 first at 1172 Elo.

Arena’s Text-to-Video Arena, last updated Nov 6, 2026, with Gemini Omni 1.2 Flash first at 1541.

Three rules keep it honest:

• A score is shown only if we read it on the board. “Not in top 25” means the model wasn’t in the first 25 rows we read, not that it has no score.
• Scores from different boards never get added or averaged. Artificial Analysis anchors its scale at Kling 3.0 1080p = 1000, and Arena’s scale is different.
• Vidwip’s models are ranked exactly like every other model. The “On Vidwip” column only marks which ones you can run.

Quick leaderboard table

Notes on the names: Arena lists Seedance as “Dreamina Seedance 2.0 (720p)” and “2.5 (720p)”, and Veo 3.1 as “Veo 3.1 (Audio)”. Artificial Analysis calls FLUX 3 Video “FLUX 3”, so we matched it by family.

Arena’s Omni 1.1 row is “Gemini Omni 1.1 Flash”; a separate unversioned “gemini-omni-flash” entry scores 1509 and isn’t counted twice.

We skipped Muse Video (Arena 1457) because we couldn’t find vendor specs for it. We also skipped Sora 2 Pro, covered in the Sora shutdown post.

What the leaderboards skip: specs and price

Price per second is Artificial Analysis’s “price per minute” divided by 60. Artificial Analysis says it is the cost on the model creator’s API at 1080p where offered, so it won’t match what you pay on a reseller.

The Gemini Omni 1.2 Flash price is Google’s own, since that model isn’t in Artificial Analysis’s top 25.

Spec cells come from vendor figures gathered in our model reviews (linked below), the SkyReels V4 paper and fal’s HappyHorse page. We couldn’t confirm them against every vendor page, so check before you budget.

1. Gemini Omni 1.2 Flash

Google’s Gemini Omni 1.2 Flash tops Arena’s text-to-video board at 1541 (±15, 1,742 votes) and its image-to-video board at 1502. Both rows are marked preliminary.

It makes 3 to 15 second clips with native audio and edits them by conversation. Our Gemini Omni Flash review covers the details, and Gemini Omni vs Veo 3.1 compares it with Google’s older model.

Pros

• Highest Arena score on both text-to-video and image-to-video
• $0.07 a second at 720p on the Gemini API

Cons

• Preliminary rows with few votes, and no Artificial Analysis score in the top 25
• Not on Vidwip

Best for: Iterating on a clip by chatting with the model. On Vidwip: No.

2. Gemini Omni 1.1 Flash

The previous Omni version scores 1514 on Arena text-to-video and 1486 on image-to-video. Artificial Analysis has it at 1127 and 1045, 12th and 16th on its boards.

That gap is the story of this post: the same model is second on one board and mid-pack on the other. We didn’t check its specs separately from 1.2.

Pros

• 1514 on Arena with a tight ±15 interval
• $0.152 a second on Artificial Analysis

Cons

• Superseded by 1.2 Flash
• Weak on Artificial Analysis (1127)

Best for: Comparing Omni versions. On Vidwip: No.

3. FLUX 3 Video

Black Forest Labs’ FLUX 3 Video is fourth on Arena text-to-video at 1491 (preliminary) and 1451 on image-to-video. Artificial Analysis lists it as “FLUX 3” at 1138 and 1071.

It makes 5 to 20 second clips with sound at 720p. Our cheapest FLUX 3 Video provider post tracks where it costs least.

Pros

• Strong Arena text-to-video score
• Up to 20 seconds with sound

Cons

• Preliminary Arena row
• $0.29 a second on Artificial Analysis, dearer than most models here

Best for: Art-directed shots up to 20 seconds. On Vidwip: Yes, 26 credits a second.

4. Grok Imagine Video 1.5

SpaceXAI’s Grok Imagine Video 1.5 scores 1486 (preliminary) on Arena text-to-video and 1454 on image-to-video. Artificial Analysis has it at 1055 and 1050.

Clips run up to 15 seconds with synced sound. Read the Grok Imagine 1.5 review for how it handles dialogue.

Pros

• Top-five on Arena text-to-video
• Native 1080p since 31 July

Cons

• Mid-table on Artificial Analysis
• No multi-shot mode

Best for: Image-to-video with lip-synced sound. On Vidwip: Yes, Grok Imagine 1.5 at 22 credits a second (Lite at 5).

5. Seedance 2.0

ByteDance’s Seedance 2.0 scores 1481 on Arena text-to-video, from 72,038 votes, and 1479 on image-to-video. Artificial Analysis has it at 1135 and 1137.

It has by far the deepest vote base in the top 10, so its score is the least likely to move. It makes up to 15 second clips with sound at 720p.

Pros

• 72,038 Arena votes behind the score
• Top 10 on all four boards

Cons

• $0.374 a second on Artificial Analysis
• 720p cap in this listing

Best for: Reliable results across both text and image prompts. On Vidwip: Yes, 30 credits a second (Fast at 19).

6. Seedance 2.5

Seedance 2.5 is within one point of 2.0 on Arena text-to-video (1480) and ahead on image-to-video (1488, third). On Artificial Analysis it is first on image-to-video at 1158 and fifth on text-to-video at 1155.

It makes takes up to 30 seconds. See our Seedance 2.5 review for limits and the 4K version still in trial.

Pros

• 30-second clips with sound
• Number one on Artificial Analysis image-to-video

Cons

• $0.569 a second on Artificial Analysis, the highest price in this table
• 720p on public APIs

Best for: Long single takes from a still image. On Vidwip: Yes, 35 credits a second.

7. Wan 3.0

Alibaba’s Wan 3.0 is number one on Artificial Analysis text-to-video at 1172 Elo (7,307 samples) and eighth on Arena at 1477, but with only 3,080 votes. Image-to-video: 1481 on Arena and 1109 on Artificial Analysis.

At $0.20 a second and up to 30 seconds with sound, it is one of the best value-for-score picks. Our Wan 3.0 review has the API pricing.

Pros

• First on Artificial Analysis text-to-video
• Up to 30 seconds at $0.20 a second

Cons

• Closed weights
• Few Arena votes, so the ±12 interval is wider than Seedance’s ±8

Best for: Long clips on a budget. On Vidwip: Yes, 16 credits a second.

8. SkyReels V4

Skywork’s SkyReels V4 sits ninth on Arena text-to-video (1458) and seventh on image-to-video (1476). Artificial Analysis has it at 1137 and 1123, at $0.35 a second.

The SkyReels-V4 paper describes a joint video and audio model reaching 1080p and 15 seconds. One third-party listing says 10 seconds for text-to-video, so we couldn’t pin the limit down.

Pros

• Top 10 on all four boards
• Joint audio and video generation

Cons

• Clip length varies between sources
• Not on Vidwip

Best for: Testing a high-ranking model outside the big labs. On Vidwip: No.

9. MiniMax H3 (Hailuo 3)

MiniMax H3 is second on Arena image-to-video at 1493 from 121,819 votes. It scores 1457 on text-to-video, and Artificial Analysis has the 768p version third at 1160.

Vidwip lists it as Hailuo 3. The Hailuo 3 review says MiniMax’s API charges $0.13 a second, and Artificial Analysis shows $0.08 at 768p.

Pros

• Second on Arena image-to-video
• Up to 2K with stereo sound

Cons

• Open-weights license excludes the US, EU, UK and South Korea
• Artificial Analysis price is for 768p, not 2K

Best for: Image-to-video at a low price. On Vidwip: Yes, as Hailuo 3 at 20 credits a second.

10. HappyHorse 1.0

Alibaba-ATH’s HappyHorse 1.0 scores 1428 on Arena text-to-video and 1443 on image-to-video. Artificial Analysis has it at 1032 and 1022, and its HappyHorse-1.1 at 1048 and 1038.

fal’s listing gives 3 to 15 seconds up to 1080p with native audio. One other site’s spec page says no audio, so we flagged it.

Pros

• Top 12 on Arena in both modes
• 1080p on fal

Cons

• Conflicting audio claims between sources
• Not on Vidwip

Best for: Alibaba-family output without Wan. On Vidwip: No.

11. Veo 3.1

Google’s Veo 3.1 scores 1363 on Arena text-to-video (rows for Audio and Audio, 1080p tie at 1363) and 1399 on image-to-video. Artificial Analysis has it at 963 and 955.

It is the oldest model in the top 12, from January 2026, and it trails Google’s own Omni. Clips are 4, 6 or 8 seconds.

See our Veo 3.1 review.

Pros

• Up to 4K on Google Cloud
• Lip-synced speech

Cons

• Short clips and $0.40 a second on Artificial Analysis
• Far behind the top 10

Best for: Dialogue scenes where its look matters. On Vidwip: Yes, 60 credits a second (Fast at 19, Lite at 8).

12. Kling 3.0

Kling 3.0 is the reference point for Artificial Analysis: the 1080p Pro model is fixed at 1000 Elo on both boards. On Arena, Kling 3.0 Pro scores 1353 on image-to-video and wasn’t in the top 25 for text-to-video.

At $0.168 a second it is cheaper than most models above it. Kling 4.0 isn’t on this list because neither board’s top 25 we read included it; our Kling 4.0 review covers it.

Pros

• Mature model at $0.168 a second
• Up to 15 seconds with sound

Cons

• Lowest-ranked of the 12
• Older release (February 2026)

Best for: Steady camera moves and product shots on a budget. On Vidwip: Yes, Kling 3.0 Standard at 20 credits a second and Pro at 26. Vidwip also runs Kling O1.

How to read these scores

A gap of a few points means nothing. Arena’s own intervals are ±8 to ±18, so Seedance 2.0 (1481) and Seedance 2.5 (1480) are a tie.

Vote counts matter too. Seedance 2.0 has 72,038 Arena votes behind its score and Wan 3.0 has 3,080, so Wan’s rank can move more.

And the boards disagree. Wan 3.0 is first on Artificial Analysis and eighth on Arena, while Gemini Omni 1.2 Flash is first on Arena and absent from Artificial Analysis’s top 25.

How to choose

• Highest blind-vote score: Gemini Omni 1.2 Flash, but it’s preliminary and not on Vidwip.
• Longest clip: Seedance 2.5 or Wan 3.0, both 30 seconds.
• Cheapest per second: Gemini Omni 1.2 Flash at $0.07 on Google’s API (720p), then MiniMax H3 at $0.08 on Artificial Analysis (768p). The bases differ.
• Best image-to-video: Seedance 2.5 on Artificial Analysis, MiniMax H3 on Arena.
• Highest resolution: MiniMax H3 at 2K, Veo 3.1 up to 4K on Google Cloud.

Want free options? See free AI video models.

Run these models on Vidwip

Eight of the 12 run in Vidwip’s AI video generator: Seedance 2.0, Seedance 2.5, Wan 3.0, Hailuo 3, FLUX 3 Video, Grok Imagine 1.5, Veo 3.1 and Kling 3.0. The full 16-model roster is on the models page, and rates are on pricing.

Credits cost $0.01 each.

AI video needs a paid plan. Starter’s 2,500 credits a month buy ten 4-second Veo 3.1 clips (240 credits each) or ten 15-second Wan 3.0 clips (240 credits each).

Final verdict

No single model wins every board, so use the ranking to build a shortlist, not to pick a winner. Gemini Omni 1.2 Flash leads Arena, Wan 3.0 leads Artificial Analysis text-to-video, and Seedance 2.5 leads Artificial Analysis image-to-video.

Scores read on 8 November 2026. Boards update often, so check the links above.