Comparisons · consumer · best chinese ai video generator

Chinese AI video, against the Sora and Veo rows we can cite

Published 16 Oct 2026. Updated 16 Oct 2026. Researched 16 Oct 2026.

People outside China usually meet Chinese video models as Kling, Hailuo, Seedance, Vidu, or Wan. Those are not one price, and a chatbot discount does not describe any of them. This page uses two files we actually stored on 16 Oct 2026: a public preference board, and vendor list prices whose units we are willing to put on the same axis.

What people preferred, on a dated board

The text-to-video file is the LMArena dataset, config text_to_video, split latest, published 22 Sep 2026. The ratings are Bradley-Terry scores, not Elo, and they are not comparable with the image board. Wan 3.0 is high and thinly voted. Kling 4.0, which Kuaishou was still describing as an October launch on 15 Oct, is not a row. Neither is a current Veo or Sora alias that post-dates this file. The chart prints the row names.

Text-to-video preference, 22 Sep 2026 snapshotdreamina-seedance-2.0-720p1479.2wan3.01476.3dreamina-seedance-2.5-720p1473.9minimax-h31460.3sora-2-pro1367.8veo-3.1-audio1364.2wan2.7-t2v1337.2wan2.6-t2v1326.7kling-2.6-pro1216.2hailuo-2.31206hunyuan-video-1.51168.8wan-v2.2-a14b1131.6
LMArena text-to-video, category overall. Bradley-Terry ratings, not Elo, and not comparable with the image or image-to-video boards. Published 2026-09-22, fetched 2026-10-16. Kling 4.0 is not a row. Source: https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset (config text_to_video, split latest).
EntryratingDetail
dreamina-seedance-2.0-720p1479.2rank 5, 56,318 votes, Proprietary
wan3.01476.3rank 6, 2,598 votes, Proprietary
dreamina-seedance-2.5-720p1473.9rank 7, 8,001 votes, Proprietary
minimax-h31460.3rank 8, 11,417 votes, minimax-h3-community-license-agreement
sora-2-pro1367.8rank 11, 54,167 votes, Proprietary
veo-3.1-audio1364.2rank 12, 13,741 votes, Proprietary
wan2.7-t2v1337.2rank 20, 27,626 votes, Proprietary
wan2.6-t2v1326.7rank 21, 54,050 votes, Proprietary
kling-2.6-pro1216.2rank 29, 73,362 votes, Proprietary
hailuo-2.31206rank 32, 89,138 votes, Proprietary
hunyuan-video-1.51168.8rank 38, 4,292 votes, tencent-hunyuan-community
wan-v2.2-a14b1131.6rank 42, 10,458 votes, Apache 2.0

Image-to-video is a different file from the same day. Vidu Q3 Pro appears here and did not appear in the text-to-video overall list we kept. MiniMax H3 is rank 1 on this split. Hailuo 2.3 is much further down, with a very large vote count, so it is a stable row rather than a missing one.

Image-to-video preference, 22 Sep 2026 snapshotminimax-h31495.4wan3.01479.8dreamina-seedance-2.5-720p1477.4dreamina-seedance-2.0-720p1474.6wan2.7-i2v1428.2veo-3.1-audio1398.1vidu-q3-pro1363.1kling-v3-pro1354hailuo-2.31262hunyuan-video-1.51197.8
LMArena image-to-video, category overall. Bradley-Terry ratings, not Elo. Published 2026-09-22, fetched 2026-10-16. Source: https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset (config image_to_video, split latest).
EntryratingDetail
minimax-h31495.4rank 1, 57,112 votes, minimax-h3-community-license-agreement
wan3.01479.8rank 3, 3,442 votes, Proprietary
dreamina-seedance-2.5-720p1477.4rank 4, 15,729 votes, Proprietary
dreamina-seedance-2.0-720p1474.6rank 5, 121,427 votes, Proprietary
wan2.7-i2v1428.2rank 11, 85,521 votes, Proprietary
veo-3.1-audio1398.1rank 13, 25,116 votes, Proprietary
vidu-q3-pro1363.1rank 18, 36,679 votes, Proprietary
kling-v3-pro1354rank 19, 206,221 votes, Proprietary
hailuo-2.31262rank 29, 405,786 votes, Proprietary
hunyuan-video-1.51197.8rank 40, 5,462 votes, tencent-hunyuan-community

Artificial Analysis publishes its own video arena. It needs an API key we do not have, so it is not a second chart. Do not average these ratings with a number from another site.

Prices that share a unit

MiniMax’s API list, fetched the same day, bills H3 per second of output: $0.08 at 768P and $0.13 at 2K. H3-Max is $0.05 at 480P and $0.08 at 768P. The older Hailuo 2.3 line is a clip price, $0.28 for a 768P, 6 second video, which is why it is not a bar below. QwenCloud’s Wan table is also per second. The wan3.0-video 1080P list price is $0.20, with a limited-time 30 percent discount printed beside it. Kling’s consumer credit grid was still not on a page we archived. See the Kling 4.0 note.

Video list prices billed per secondMiniMax-H3-Max 480P0.05MiniMax-H3 768P0.08wan2.7-t2v 720P0.1MiniMax-H3 2K0.13wan2.7-t2v 1080P0.15wan3.0-video 1080P list0.2
List prices billed per second of output. Kling consumer credits are not included: no official grid was archived. QwenCloud's wan3.0-video row is the list price; the same page also shows a limited-time 30% discount. Fetched 2026-10-16.
EntryUSD per secondDetail
MiniMax-H3-Max 480P0.05platform.minimax.io video list
MiniMax-H3 768P0.08platform.minimax.io video list
wan2.7-t2v 720P0.1QwenCloud model-prices page
MiniMax-H3 2K0.13platform.minimax.io video list
wan2.7-t2v 1080P0.15QwenCloud model-prices page
wan3.0-video 1080P list0.2QwenCloud list; sale price shown as $0.14

Seedance is a token price, on BytePlus

BytePlus ModelArk is the international console for ByteDance’s video models. Dreamina Seedance 2.x is billed per million tokens, with a lower rate when the input includes video. That cannot sit on the per-second chart without a conversion we are not going to invent. The BytePlus guide has the signup and payment limits. Seedance 2.5 also has prepaid token packs on the offers page, from $32 for 5 million tokens, valid three months, and those packs are not refundable once bought.

Seedance 2.x, input without video, per million tokensSeedance 2.0 mini3.5Seedance 2.0 fast5.6Seedance 2.0, 480p/720p7Seedance 2.0, 1080p7.7Seedance 2.510.7
Dreamina Seedance 2.x rates labelled input without video on the ModelArk product page, fetched 2026-10-16. The with-video rate is lower and is in the extra column. This is not a price per second, and it is not comparable with the per-second video chart. Source: https://www.byteplus.com/en/product/ModelArk
EntryUSD per 1M tokens, input without videoDetail
Seedance 2.0 mini3.5with video input $2.1 / M
Seedance 2.0 fast5.6with video input $3.3 / M
Seedance 2.0, 480p/720p7with video input $4.3 / M
Seedance 2.0, 1080p7.7with video input $4.7 / M
Seedance 2.510.7with video input $6.4 / M

What we will not rank

A “best generator” sentence needs a prompt you care about. The board is blind preference on someone else’s prompts, frozen on 22 Sep. Kling 4.0’s feature table is real and its credit price is not in this repo. Vidu has a QwenCloud per-second line for some ids and a preference row only on image-to-video. Jimeng is the China consumer name people use for Dreamina-class tools. We did not open a China app store.

Privacy

A video prompt plus a face is a biometric upload. Kling is Kuaishou, Hailuo is MiniMax, Seedance is ByteDance, Wan is Alibaba. All of those companies are in the PRC even when the console you pay is an international cloud. Do not upload a face you could not explain. The safety guide is the longer version. Open weights, if you want the clip on your own GPU, are in the local video guide.

Related: Kling, Hailuo, Kling 4.0, BytePlus, free video and image.

Questions

Which Chinese video model is ahead of Sora and Veo?

On the 22 Sep 2026 text-to-video snapshot, Seedance, Wan, and MiniMax H3 sit above the Sora and Veo rows we stored. That file has no Kling 4.0 row.

Is Seedance cheaper per second than Hailuo?

We cannot say that. BytePlus prices Seedance per million tokens. MiniMax prices H3 per second. Those units are not the same chart.

Sources

  1. LMArena leaderboard dataset, accessed 2026-10-16
  2. MiniMax model pricing, accessed 2026-10-16
  3. QwenCloud model prices, accessed 2026-10-16
  4. BytePlus ModelArk, accessed 2026-10-16
  5. Kling 4.0 user guide, accessed 2026-10-16