Skip to content
trending

AI Tools

Seedance 2.5: Versions, Pricing, Rankings and Test Results

Seedance 2.5 ranks third in Artificial Analysis text-to-video, but our Polish-language tests exposed weak speech and text rendering.

By 11 min read
Seedance 2.5: Versions, Pricing, Rankings and Test Results

👁 141 reads

Updated regularly. Information current as of October 11, 2026 (previous version: September 27).

Seedance is ByteDance’s family of video models (ByteDance also owns TikTok and CapCut). They create clips from text, images and references, with sound generated alongside the video. The latest model, Seedance 2.5, ranks third in Artificial Analysis’s text-to-video leaderboard (1,143 Elo) and can generate a 30-second clip in one pass. Seedance 2.0 ranks fifth in image-to-video. Seedance charges by video token, so below I explain how to estimate the cost in dollars and share results from my tests: eight clips across four rounds, including Polish speech and on-screen text. We tested the models in Polish; the English renderings of Polish test material below are translations, not new tests.

Status as of October 11, 2026

  • Latest version: Seedance 2.5 (Dreamina Seedance 2.5), launched July 31, 2026, in Dreamina, Jimeng and Doubao; available through the BytePlus API from August 6.
  • Artificial Analysis ranking: Seedance 2.5 ranks third in text-to-video (1,143 Elo); Seedance 2.0 ranks seventh in text-to-video (1,116) and fifth in 720p image-to-video (1,176).
  • Price according to AA: Seedance 2.5 costs $34.12 per minute, the highest price in the top 10; Seedance 2.0 costs $22.45.
  • Where to use it: Dreamina, BytePlus ModelArk, OpenRouter, fal.ai, Higgsfield, Runway and other platforms.
  • Polish: Polish is not among Seedance 2.5’s 11 officially listed languages. In my Polish speech test, Seedance 2.5 got every other word wrong.
  • My tests: 2.0 Fast scored 7.5/10 (September 27), 2.5 in the Arena scored 5.6/10 (October 8), and 2.5 image-to-video scored 7.0/10 (October 11).
3.Seedance 2.5’s text-to-video ranking in AA
30 slongest Seedance 2.5 clip in one pass
50references at once: 30 images, 10 videos, 10 audio files
10.70USD per million Seedance 2.5 tokens on OpenRouter
Official Dreamina Seedance 2.5 launch video (Dreamina’s YouTube channel, July 31, 2026).

What is Seedance?

The Seed team at ByteDance develops Seedance. The company distributes the model through three channels:

  • ByteDance apps - international app Dreamina (part of the CapCut family), plus Jimeng AI and Doubao in China. That is why leaderboards call the model “Dreamina Seedance.”
  • BytePlus ModelArk - ByteDance’s official API outside China, excluding the US, with pay-as-you-go token pricing or packages.
  • Third-party platforms - OpenRouter, fal.ai, Higgsfield, Runway, Magnific and others, which buy access and resell it through their own credits.

Seedance stands out for two things: it generates audio alongside video, and it supports references. You can provide characters, products, motion recordings and voice samples for a clip, and the model is designed to keep them consistent throughout the scene.

Compare all the generators: AI video generator ranking, October 2026 prices and Polish-language test · For Polish-language use without a subscription: ZWID (my service)

Seedance 2.5: the latest version

ByteDance announced Seedance 2.5 at the FORCE conference on June 23, 2026, and released it on July 31, 2026. According to a post from the Seed team, the main changes from version 2.0 are:

  • 30 seconds in one pass instead of 15, with multiple shots and the option to extend a clip repeatedly. The company says this can produce films several minutes long.
  • Up to 50 references at once: 30 images, 10 video clips and 10 audio recordings, including 3D “clay renders” to guide camera placement and movement.
  • Editing down to the second: revise a selected section, such as a character, action or camera move, while leaving the rest of the clip unchanged.
  • Green-screen editing and camera-perspective changes in finished footage.

Resolution varies by platform: OpenRouter lists 480p and 720p for version 2.5, while fal.ai’s schema also shows 1080p, and Dreamina advertises “4K up to 30 seconds.” I tested only 480p through the API. Dreamina’s official list of supported speech languages includes Chinese, English, Spanish, Indonesian, Malay, Thai, Arabic, Portuguese, Vietnamese, Japanese and Korean. See Seedance 2.5 for more on version 2.5 and “unlimited” offers, and What Seedance 2.5 can do for examples and prompts.

Artificial Analysis rankings

Ranking Model Place Elo Price per minute according to AA
Text-to-video Dreamina Seedance 2.5 3 of 29 1,143 $34.12
Text-to-video Dreamina Seedance 2.0 7 of 29 1,116 $22.45
Image-to-video Dreamina Seedance 2.0 720p 5 of 36 1,176 $9.07
Image-to-video Seedance 1.5 Pro 28 of 36 1,000 (reference score) $11.86

Seedance 2.5 does not appear in the image-to-video table. These are readings from AA-Video-T2V v2.0 and AA-Video-I2V v1.0 on October 11, 2026.

Text-to-video Elo rankings (AA, October 11, 2026)
Wan 3.01,156
Utopai X1,149
Seedance 2.51,143
MiniMax H3 (768p)1,137
FLUX 31,128
Seedance 2.01,116

The scale starts at 1,000 points. In an AA category analysis dated September 30, Seedance 2.5 was narrowly behind Utopai X in physics. See Seedance 2.5 vs. Wan 3.0 vs. Utopai X for a comparison.

Seedance versions and API pricing

Figures are from OpenRouter model cards (September 27; rates checked again on October 11, 2026). Prices are in US dollars, exclude VAT and are per million video tokens for generation without an input video.

Version Duration Resolution Price per 1 million tokens 5-second 720p clip
Seedance 1.5 Pro 4-12 s 480p-1080p $2.40 with audio, $1.20 without $0.26
Seedance 2.0 4-15 s 480p-4K $7.00 (1080p: $7.70, 4K: $4.00) $0.76
Seedance 2.0 Fast 4-15 s 480p, 720p $4.20 $0.46
Seedance 2.0 Mini 4-15 s 480p, 720p $3.50 $0.38
Seedance 2.5 4-30 s 480p, 720p $10.70 $1.17

Seedance 1.5 Pro has been available on OpenRouter since March 2026; versions 2.0 and 2.0 Fast arrived in April, and Mini and 2.5 in August.

How much does Seedance cost? Tokens, API and Dreamina

Seedance does not charge per second. You pay for tokens, and the number of tokens depends on resolution and frame count:

  • frames = 24 × duration in seconds + 1,
  • tokens = width × height × frames / 1024,
  • cost = tokens × price per million / 1,000,000.

I checked this formula against my own bills. A 4-second clip at 1280×720 has 97 frames, so 1280 × 720 × 97 / 1024 = 87,300 tokens; at $4.20 per million, that comes to $0.36666 - exactly what OpenRouter returned in the usage.cost field for Seedance 2.0 Fast. A Seedance 2.5 clip at 854×480 with 121 frames uses 48,440 tokens and costs $0.5183, again matching the bill exactly. What this means:

  • 1080p costs more than twice as much as 720p because it has 2.25 times as many pixels.
  • 480p costs less than half as much as 720p: a 30-second Seedance 2.5 clip costs about $3.09 at 480p and about $6.94 at 720p on OpenRouter.
  • The $34.12-per-minute price in the AA table is much higher than a minute of 720p on OpenRouter (about $13.88). AA does not say which configuration it uses.
Provider Seedance 2.5 rate 5 seconds at 480p
OpenRouter $10.70 per million tokens $0.52 (my calculation)
fal.ai $21.40 per million tokens ($0.0214 per thousand) About $1.04; my 4:3 clip cost $1.06
BytePlus ModelArk Rates and packages listed in the BytePlus pricing schedule (not available in the US) -

Dreamina uses subscription plans and credits. The plans and free-credit offer vary by market; check Dreamina for current availability and pricing. See Dreamina: pricing and free credits for details and promotions. For a comparison of model rates, see How much does a second of AI video cost?

Dealat Kinetyra
Runway Pro - 12 months
$159 instead of $373 -57%
See the deal →

My Seedance tests

Seedance 2.5 in the video Arena (October 8)

Seedance 2.5 received the same five prompts as Wan 3.0, MiniMax H3 Max, Kling 3.0, Veo 3.1 Lite and Runway Gen-4.5: 5 seconds, 480p, 16:9, one attempt, through OpenRouter ($0.52 per clip). The Polish speech result came from the September 28 round, using the same prompt. English renderings of the Polish test material are translations of the original prompts, not prompts tested in English.

Scene Seedance 2.5 Best in scene Result
Orbit around a motorcycle 6.0 Wan 3.0, Kling 3.0, FLUX 3 - 8.5 The orbit is good, but a black sheet appears on the seat.
Polish speech 4.0 FLUX 3, H3 Max - 9.0 WER 0.583 - every other word is wrong. The test was conducted in Polish.
On-screen text, translated from the Polish original: “YELLOW BOAT” 5.0 Wan 3.0, H3 Max, FLUX 3 - 10 The original Polish text rendered with corrupted lettering in the original Polish test, mixing German and Czech characters.
Water physics 7.0 Wan 3.0 - 9.0 The glass fills too quickly.
Character in two shots 6.0 Wan 3.0 - 9.0 The hood hides the character’s hair, and the backpack is not visible.
Average 5.6 FLUX 3 - 8.9 Fifth of seven models
My Seedance 2.5 video Arena clip: water being poured into a glass, 480p, 7.0/10.

Seedance 2.5 image-to-video (October 11)

I gave Seedance 2.5 and FLUX 3 the same café photo and the same prompt on fal.ai. Seedance made a gentle camera move and, as requested, showed a woman at a table drinking coffee. However, the 480p image was soft and one person at the table disappeared: 7.0/10 at a cost of $1.06. FLUX 3 scored 8.5/10 and cost $0.425. Both clips are in the comparison FLUX 3 vs. Wan 3.0 vs. Seedance 2.5.

Seedance 2.0 Fast (September 27)

On September 27, 2026, I generated two clips through OpenRouter using the same prompts given to Veo 3.1 Fast, Kling 3.0, Hailuo 2.3, Grok Imagine Video and Wan 3.0. Settings: 4 seconds, 720p, 16:9, audio enabled, fixed random seed and one attempt. See the full comparison in the AI video generator ranking. The examples below translate the original Polish test prompts; they were not retested in English.

Seedance 2.0 Fast, coffee, 8.0/10: an empty cup fills in front of the viewer, with a clear camera push-in and a kitchen with a window. Downsides: a rainbow flare and added music.
Seedance 2.0 Fast, tram, 7.0/10: a tram travels sideways across the entire frame from a stationary camera. It is cream and red, in the style of a Prague Tatra, but the umbrellas are barely visible.
Coffee Tram
Tokens 87,300 87,300
Cost (usage.cost) $0.36666 $0.36666
Time until file was available to download 164 s 106 s
File 1280×720, 4.04 s, with audio 1280×720, 4.04 s, with audio
Audio Pouring and a melody (notes D, E, F, C) Street noise, no speech
Score 8.0/10 7.0/10

One audio-analysis oddity: faster-whisper with speech detection treated 3.9 seconds of the coffee clip’s soundtrack as speech and “transcribed” it as “Thanks for watching!” That is a typical Whisper hallucination; the audio spectrum shows a melody, not speech. Seedance added music that the prompt did not request.

Does Seedance support Polish?

  • Not officially: Polish is not among the 11 languages listed for Seedance 2.5 in Dreamina.
  • Speech in my test: Seedance 2.5 had a WER of 0.583 on a Polish sentence about a beetle in Szczebrzeszyn - more than every other word was wrong. In a test using text about Dreamina, Seedance 2.0 Mini mispronounced two of five words. These were Polish-language tests.
  • On-screen text:the tested Polish phrase (translated from the Polish original as “YELLOW BOAT”) came out with corrupted lettering in the original Polish test. The English wording here is a translation of the tested original, not a new test.
  • What to do: for Polish dialogue and captions, Wan 3.0, FLUX 3 Video and MiniMax H3 Max performed better in my tests. It is better to add speech separately when using Seedance.

How to get started with Seedance

  1. Without coding: go to dreamina.capcut.com, sign in (free accounts get a daily credit allowance) and choose Dreamina Seedance 2.5 or 2.0 in the video generator.
  2. Write your prompt in English or add references such as a character or product image, a motion recording or a voice sample.
  3. Start with a short clip and low resolution to avoid spending credits on unsuccessful attempts.
  4. Through an API: create an account with OpenRouter or fal.ai, add funds and call bytedance/seedance-2.5 (OpenRouter) or bytedance/seedance-2.5/text-to-video (fal.ai). The per-token rate on fal.ai is twice as high.
  5. For longer scenes, use 480p and render only selected shots at higher quality.

Seedance compared with the competition

Model T2V Elo USD per minute (AA) Longest clip My Arena score
Wan 3.0 1,156 $12.00 30 s 8.8
Utopai X 1,149 No API Not specified No API
Seedance 2.5 1,143 $34.12 30 s 5.6
MiniMax H3 Max 1,130 $4.80 15 s 8.4
FLUX 3 Video 1,128 $17.40 20 s 8.9
Seedance 2.0 1,116 $22.45 15 s 2.0 Fast: 7.5 (different round)

For more, see AI video generators: which one should you choose?, the video Arena and Vidu Q4 vs. Seedance 2.0 vs. MiniMax H3 for image-to-video.

Who is Seedance for?

  • Scenes with multiple references - characters, products and voices in one clip; its 50 input files are more than any other model on the market.
  • Long, story-driven shots - 30 seconds with multiple shots in one pass.
  • Dreamina and CapCut users - subscription plans and free credits, with no API required.
  • Not for Polish dialogue and captions - that was the weakest area in my tests.
  • Not for the cheapest batches - through the API, Seedance 2.5 costs several times more than H3 Max and Wan 3.0.

My view: Seedance 2.0 Fast makes sense if you need references (a face, product or voice) or clips longer than 8 seconds that Veo cannot make. I would choose Seedance 2.5 for complex scenes prompted in English, and FLUX 3 or Wan 3.0 for individual Polish-language shots.

If you want Kling, Seedance, Veo and image models in one account, see my Higgsfield guide for pricing details.

Update: 17 November 2026

The art director approves a photo. The choreographer approves movement in a video. Seedance can take different kinds of references, and on fal’s Seedance 2.5 Flash reference route those inputs affect the arithmetic. The form accepts images, video and audio; the published billing formula includes input-video and output-video duration.

For a generation without a video reference the fal estimate lists roughly $0.1442 per second at 480p and $0.3094 at 720p. With video input it publishes a separate multiplier and accounts for reference duration. I would avoid carrying the simple no-video rate into a shot that relies on a movement reference.

Label each authorized asset by its job. Use a shopkeeper’s approved product photograph for appearance and an original hand movement for timing. Keep the audio setting and requested duration alongside the final quote so the editor can reproduce the conditions.

Practical details

The fal form provides an audio reference field as well as a separate generated-audio setting. Record both when they matter to the brief.

Sources

Frequently asked questions

How much does Seedance cost?

On OpenRouter, rates range from $1.20 to $10.70 per million video tokens, depending on the version. In practice, a 5-second 720p clip costs from $0.26 (1.5 Pro) to $1.17 (2.5). My Seedance 2.5 clip (5 seconds, 480p) cost $0.52.

How do you calculate Seedance tokens?

Multiply the width by the height and the number of frames (24 per second plus one), then divide by 1024. A 4-second 1280×720 clip uses 87,300 tokens, which my bill confirmed to the cent.

Is Seedance free?

Dreamina provides free credits each day, with a watermark on files. For regular use, you need a paid plan or a funded API account.

Which Seedance version is best?

Seedance 2.5 ranks highest in AA’s text-to-video leaderboard (third place), while Seedance 2.0 ranks fifth in image-to-video. Choose 2.0 for 1080p and 4K through the API, 2.5 for longer scenes, and 2.0 Mini or 1.5 Pro for lower-cost batches.

Does Seedance speak Polish?

Polish is not officially supported, and in my test Seedance 2.5 got more than half the words in a Polish sentence wrong. Wan 3.0, FLUX 3 and H3 Max performed better for Polish dialogue.

What is the difference between Seedance 2.5 and 2.0?

Version 2.5 makes 30-second clips instead of 15-second clips, accepts 50 references and lets you edit a selected section. On OpenRouter it offers only 480p and 720p, while version 2.0 goes up to 4K. The 2.5 rate is about 50% higher: $10.70 versus $7.00 per million tokens.

Where can you use Seedance besides Dreamina?

It is available through BytePlus ModelArk (the official API, not available in the US), OpenRouter, fal.ai and creator platforms including Higgsfield and Runway.

What does “Dreamina Seedance” mean in the leaderboard?

It is the Seedance model. ByteDance uses that name because Dreamina was the first of its apps to offer it.

More about Seedance

Google · Your sourcesQuantslant higher in your Google - one clickAdd to preferred sources →

Sources and date checked

Official sources

The link to Higgsfield is a partner link: if you buy through it, QUANTSLANT earns a commission and you pay the same price. It has no influence on the content, including where the coverage is critical.

// read next

More on this from QUANTSLANT

Piotr Olszewski

Editorial team

Independent AI publication. Models and tools are tested on real tasks, prices are checked against official rate cards, and articles are updated when things change.

read nextGoogle Veo 3: Pricing, Versions & Test Resultsai toolsGoogle Veo 3: Pricing, Versions & Test ResultsGoogle Veo 3.1 comes in Quality, Fast and Lite versions. We tested Veo 3.1 Fast against five competitors - here's what it costs and how it performed.

Most read

  1. 1Best AI Video Generator Apps for Your Phone (2026)
  2. 2Claude Opus 5.5 Motion Graphics: 60s Showreel from One Prompt
  3. 3Wan 2.2 on RTX 4090: timings, VRAM and four test clips
// site map

This page in other languages6

Quantslant in numbers 1,544posts in archive11news from the last 7 days18video models in the observatory
🎬 AI Video Benchmarks: Higgsfield DoP vs OpenAI Sora • Higgsfield vs Runway Gen-3 • AI Video Credit Calculator
× ‹ enlarge ›