Utopai X vs Wan 3.0 vs MiniMax H3: rankings, prices and access
Compare Utopai X, Wan 3.0 and MiniMax H3 on video rankings, API access, PAI costs, clip features and clearly attributed source tests.
👁 113 reads
Utopai X, Wan 3.0 and MiniMax H3 all appear in the top five of the Artificial Analysis text-to-video ranking checked on October 11, 2026. Two are related: Utopai X is fine-tuned from MiniMax H3. This comparison covers Elo scores, per-minute costs, clip duration and access, alongside the source author’s earlier tests of Wan 3.0 and H3 Max. Those tests were not repeated for this English adaptation. For the full studio-model guide, see Utopai X.
At a glance
- Highest ranked: Wan 3.0 at 1156 Elo, though its confidence interval overlaps Utopai X at 1149.
- Lowest listed cost: MiniMax H3 at $4.80 per minute in AA’s table, with API pricing measured in cents per second.
- Highest calculated cost: Utopai X has no public API. Its PAI credit rate works out at about $60-$84 per minute of 1080p video, depending on the subscription.
- Source language tests: Wan 3.0 and H3 Max reproduced the tested Polish speech accurately. Utopai X was not tested. These are Polish-language observations, not evidence of English voice quality.
- Practical shortlist: Consider Wan 3.0 for API workflows and automation, H3 for lower-cost batches, and Utopai X for teams already working inside PAI.
Artificial Analysis rankings
| Model | T2V rank | T2V Elo and 95% confidence interval | Comparisons | I2V rank and Elo | Release month listed by AA |
|---|---|---|---|---|---|
| Wan 3.0 | 1 | 1156 (1147-1165) | 8300 | 7 (1164) | August 2026 |
| Utopai X | 2 | 1149 (1139-1159) | 6071 | Not listed | September 2026 |
| MiniMax H3 (768p) | 4 | 1137 (1128-1146) | 8315 | 2 (1181) | July 2026 |
| MiniMax H3 Max | 5 | 1130 (1120-1140) | 5719 | 1 (1195) | August 2026 |
Snapshot from October 11, 2026: text-to-video, AA-Video-T2V v2.0 and image-to-video, AA-Video-I2V v1.0. The MiniMax family leads this image-to-video snapshot, while Utopai X ranks above both H3 variants in text-to-video. These are different evaluation categories.
The chart starts at 1000 Elo. Utopai X scores 12 points above the base H3 entry at 768p. That is the observed ranking difference, rather than a controlled measurement isolating the effect of fine-tuning.
How Utopai X relates to MiniMax H3
The source guide dates MiniMax H3’s release to July 31, 2026 and describes H3-Base as an open-weight release. It also flags a territorial restriction in that release’s license; anyone planning to deploy the weights should check the actual license for their location and use case. Related fine-tuned models include H3 Max, attributed to fal.ai and oriented toward speed and prompt adherence, and Utopai X, adapted by the film studio for physics, camera work, lighting and audio. Artificial Analysis explicitly identifies MiniMax H3 as Utopai X’s base.
The practical distinction is access. An open-weight base does not imply that every fine-tuned derivative is available on the same terms. At the source check, Utopai X offered neither downloadable weights nor a public API; access was through PAI.
Per-minute costs
| Model | AA price per minute | Provider pricing or access | Source author’s five-second Arena cost |
|---|---|---|---|
| Wan 3.0 | $12.00 | $0.05 / $0.10 / $0.20 per second at 480p / 720p / 1080p | $0.21 at 480p |
| Utopai X | No API | 93 PAI credits per second at 1080p; equivalent to about $60-$84 per minute | Not tested |
| MiniMax H3 (768p) | $4.80 | Available through the Hailuo app and MiniMax API | Not tested in the source Arena |
| MiniMax H3 Max | $4.80 | $0.08 per second at 768p; the source lists this for OpenRouter and as the fal.ai rate after October 15 | $0.25 at 480p |
Wan pricing is documented in the Wan 3.0 guideand H3 Max pricing in the MiniMax H3 Max guide. PAI pricing was checked for the source article on October 11. At the Basic plan’s credit rate, one minute of Utopai X has an equivalent cost of roughly 17 minutes of H3 or seven minutes of Wan 3.0 at the cited AA rates. This is a unit-cost comparison, not a claim that one Basic subscription allowance covers a whole minute. Provider, resolution and billing terms differ, and recorded test charges can differ from current list rates.
Features
| Feature | Wan 3.0 | Utopai X | MiniMax H3 / H3 Max |
|---|---|---|---|
| Maximum clip duration | 30 seconds in one generation | Not specified | 5-15 seconds for H3 Max |
| Resolution | 480p, 720p, 1080p | 1080p rate in PAI; the platform supports exports up to 4K | 480p and 768p; 1080p through a refinement stage |
| Audio | Yes | Yes | Yes, but the cited OpenRouter H3 Max API route outputs no audio |
| Inputs | Text, images, audio, video and documents such as PDF and DOC | Text plus project context in PAI | Text, first and last frames, and references |
| API | Yes: Alibaba Cloud, fal.ai and OpenRouter | No public API | Yes: MiniMax, fal.ai and OpenRouter, depending on the variant |
| Polish-language source tests | Accurate dialogue in the tested scene | Not tested | H3 Max reproduced the tested tongue twister accurately |
AA categories: where each model leads
Utopai’s account of the September 30 Artificial Analysis breakdown reports:
- Physics - Utopai X first, narrowly ahead of Seedance 2.5.
- Audio synchronization - Utopai X first, narrowly ahead of Gemini Omni Flash 1.1.
- Lighting and materials - Wan 3.0 first, Utopai X second.
- Camera work - Wan 3.0 first, Utopai X second.
- Compared with base H3 - Utopai X narrowed the gap to the category leader in seven of ten categories, most notably audio, camera work and physics.
The published side-by-side comparisons of Utopai X, Wan 3.0, Seedance 2.5 and MiniMax H3 are embedded in What Utopai X can do: six examples.
The source author’s Wan 3.0 and H3 Max tests
In the Video Arena on October 8, the author gave both models the same five scenes: five seconds, 16:9, one attempt per scene. Wan 3.0 averaged 8.8/10 and H3 Max 8.4/10. Wan scored higher on water physics, 9.0 versus 7.5, and character consistency, 9.0 versus 7.5. H3 Max scored higher on Polish speech, 9.0 versus 7.5. Both reproduced the requested Polish lettering accurately. Utopai X was not included because the test protocol requires API access. The scores describe that particular mixed test, including its Polish-language tasks; they do not establish a ranking for English dialogue.
Which model fits your workflow?
- Wan 3.0 - consider it for API access, shots up to 30 seconds and document-based video generation. Its position in this ranking and its lower listed cost than Utopai X make it a useful starting point for comparison.
- MiniMax H3 / H3 Max - consider these when cost matters. They have the lowest listed per-minute price among this top-five group and occupy the first two places in the cited image-to-video ranking. Choose the variant and provider carefully, especially if you need audio.
- Utopai X - consider it when producing a film or advertisement inside PAI and using scripts and previous shots as context. For an independent automated pipeline, the lack of a public API is a material limitation.
Related guides: AI video generators: how to choose, how much a second of AI video costs and FLUX 3 Video.
Frequently asked questions
Is Utopai X simply MiniMax H3?
It is a fine-tuned version developed by Utopai Studios. In this AA snapshot it scores 12 Elo points above the base H3 entry at 768p. It also has different access and pricing arrangements.
Which of the three is cheapest?
MiniMax H3 is listed at $4.80 per minute in AA’s table. Wan 3.0 is $12, while Utopai X has no public API price and instead uses PAI subscription credits.
What do the speech tests tell us?
The source author observed accurate Polish dialogue from Wan 3.0 and an accurate Polish tongue twister from H3 Max. Utopai X was not tested because it lacked API access. No equivalent English speech comparison is established by those results.
Sources and verification date
- Artificial Analysis T2V and I2V (October 11
- 2026)
- Utopai: AA category analysis for Utopai X
- PAI pricing
- AA on the Utopai X launch. Source author’s tests: Video Arena (October 8
- 2026)
- Wan 3.0
- MiniMax H3 Max. Source checks dated October 11
- 2026. This English adaptation does not claim a new hands-on test


