Seedance 2.5 vs Wan 3.0 vs Utopai X: prices and video results
Compare three leading AI video models by AA scores, API access, pricing, references and source test results, with clear language-test limitations.
👁 112 reads
As of October 11, 2026, Wan 3.0, Utopai X and Seedance 2.5 occupy the top three places in the Artificial Analysis text-to-video ranking, separated by 13 Elo points. Wan and Seedance have API access; Utopai X runs only inside its studio’s PAI platform. This comparison covers ranking results, costs, features and the source author’s tests. Those tests include Polish speech and lettering, so their overall scores should not be treated as English-language benchmark results. For the full ByteDance model guide, see Seedance.
At a glance
- AA ranking: Wan 3.0 scores 1156 Elo, Utopai X 1149 and Seedance 2.5 1143. Their confidence intervals overlap.
- Cost per minute: Wan 3.0: $12 in AA’s table. Seedance 2.5: $34.12 in AA’s table. Utopai X: an equivalent of about $60-$84 through PAI plans.
- Duration: Wan 3.0 and Seedance 2.5 support 30 seconds in a single generation. The cited Utopai X documentation does not specify a maximum.
- References: Seedance 2.5 supports up to 50 reference files, the largest stated allowance in this comparison.
- Language-test limitation: Wan 3.0 handled the source author’s Polish speech and lettering tests well; Seedance 2.5 did not. Utopai X was not tested under that API-based protocol. These findings do not establish English speech quality.
Ranking and category results
| Model | Company | T2V Elo and confidence interval | Comparisons | AA categories, September 30 |
|---|---|---|---|---|
| Wan 3.0 | Alibaba | 1156 (1147-1165) | 8300 | First for lighting and materials, and for camera work |
| Utopai X | Utopai Studios | 1149 (1139-1159) | 6071 | First for physics and audio synchronization |
| Seedance 2.5 | ByteDance | 1143 (1134-1152) | 8264 | Second for physics, narrowly behind Utopai X |
Ranking: AA-Video-T2V v2.0, October 11, 2026. Category results are from AA’s analysis as reported by Utopai Studios. The three confidence intervals overlap, so the displayed order should not be mistaken for a decisive performance gap.
The same scene in the AA Arena
All three models, plus MiniMax H3, appear in the Artificial Analysis Arena comparisons published by Utopai. This physics example shows a character pushing a boat into water:
The other five comparisons cover lighting, construction, a robot with audio, presenters and camera movement. They are collected in What Utopai X can do.
Pricing
| Model | AA price per minute | Billing and access | Source author’s five-second test cost |
|---|---|---|---|
| Wan 3.0 | $12.00 | API: $0.05 / $0.10 / $0.20 per second at 480p / 720p / 1080p | $0.21 at 480p |
| Seedance 2.5 | $34.12 | Token-based API billing: OpenRouter $10.70 per million tokens; fal.ai $21.40. Dreamina uses a separate subscription, with local checkout pricing to be checked. | $0.52 at 480p through OpenRouter |
| Utopai X | No API | PAI: 93 credits per second at 1080p; plans from $15 to $379 per month | Not tested |
Using the calculation in the source guide, Seedance 2.5 at 720p through OpenRouter works out at about $13.88 per minute. That is substantially below AA’s $34.12 figure, showing why provider and output settings matter. It is a configuration-specific estimate rather than a universal price. The formula is in the Seedance guide.
Features
| Feature | Wan 3.0 | Seedance 2.5 | Utopai X |
|---|---|---|---|
| Maximum clip duration | 30 s | 30 seconds, with extensions to several minutes | Not specified |
| References | Images, audio, video and documents; one document of up to 50 pages | Up to 50 files: 30 images, 10 videos, 10 audio files; also supports 3D mockups | Project context within PAI |
| Editing | Not verified in the source test | Timestamp-based changes, green-screen output and camera changes | PAI editor |
| Resolution | 480p, 720p, 1080p | 480p and 720p through OpenRouter; 1080p on fal.ai | 1080p generation rate; PAI exports up to 4K |
| API | Yes | Yes | No |
| Polish-language source test | Accurate dialogue and lettering in the tested scenes | Word error rate 0.583; the requested lettering was corrupted | Not tested |
What the source author tested
In the Video Arena test on October 8, all models received the same five prompts at 480p. Wan 3.0 scored 8.8/10 and Seedance 2.5 scored 5.6/10. Polish speech and lettering accounted for much of the difference, although Wan also scored higher on water physics, 9.0 versus 7.0, and character consistency, 9.0 versus 6.0. AA uses a different evaluation and its overall ranking separates the two by only 13 Elo points. Utopai X was absent from the source author’s test because it had no public API. The clips below reproduce that earlier test, not a new English-language experiment.
How to choose
- Wan 3.0 - a practical starting point if you want API access and a lower listed cost. It leads this AA snapshot and performed well in the source author’s tests, including document-based generation and Polish-language scenes.
- Seedance 2.5 - worth considering when a scene depends on many character, product or voice references, or when extending and editing existing clips matters. Test dialogue in your target language before relying on it; the poor Polish result does not automatically apply to English.
- Utopai X - most relevant to teams already producing work in PAI and looking for a model that uses script and project context. Its lack of a public API limits standalone integrations.
Related comparisons: Utopai X vs Wan 3.0 vs MiniMax H3, FLUX 3 vs Wan 3.0 vs Seedance 2.5, Vidu Q4 vs Seedance 2.0 vs MiniMax H3.
Frequently asked questions
Is Seedance 2.5 better than Wan 3.0?
It scores 13 Elo points lower in this AA snapshot, with overlapping confidence intervals. It also scored lower in the source author’s mixed-scene test, 5.6 versus 8.8, which included Polish-language tasks. Its stronger reasons for consideration are its stated reference allowance and clip-editing features.
Which of the three costs least?
Wan 3.0 has the lowest cited rate: $12 per minute in AA’s table, or $0.05 per second at 480p in the provider pricing used for this comparison. Resolution, provider and billing rules can change the actual charge.
What does the language test establish?
Wan 3.0 handled the tested Polish dialogue well. Seedance 2.5 did not list Polish among its supported languages and misrecognized more than half the words in that test. Utopai X was not tested. These results should not be repackaged as a comparison of English voice quality.
Sources and verification date
- Artificial Analysis T2V (October 11
- 2026)
- Utopai: AA category results and comparisons
- PAI pricing
- ByteDance Seed: Seedance 2.5
- OpenRouter Seedance 2.5. Source author’s tests: Video Arena
- Wan 3.0. Source checks dated October 11
- 2026. This adaptation preserves their date and does not claim a new hands-on test
