HiDream-O1-Video vs MiniMax H3 Max vs Vidu Q4 vs Seedance 2.0
Elo ratings, cost per minute, clip length, resolution, and availability. A detailed comparison of HiDream-O1-Video, MiniMax H3 Max, Vidu Q4, and Seedance 2.0.
👁 117 reads
HiDream-O1-Video-1.0, Vidu Q4 Preview, Seedance 2.0, and MiniMax H3 Max are separated by only 20 Elo points (1175-1195) on the Artificial Analysis image-to-video leaderboard. Outside of H3 Max, these score gaps fall within the statistical margin of error. As a result, your choice comes down to pricing, clip length, generation modes, and access methods. HiDream is inexpensive ($5.80 per minute), but it features the narrowest API and lacks third-party aggregator support. Full model profile: HiDream-O1-Video.
Verdict in Brief (October 11, 2026)
- Best value for money: MiniMax H3 Max - 1st place (1195) and the lowest price at $4.80 per minute.
- Highest resolution and reference control: Vidu Q4 Preview - up to 4K, up to 15 reference images, $5.04 per minute.
- Broadest availability and text-to-video: Seedance 2.0 - BytePlus, OpenRouter, fal.ai, but the most expensive ($9.07).
- HiDream-O1-Video: image-to-video only via HiDream’s own API; best suited for physics-heavy scenes and model-determined clip durations.
Rankings: Who Actually Comes Out Ahead
| Model | I2V Rank | Elo | 95% CI | Comparisons | T2V Rank |
|---|---|---|---|---|---|
| MiniMax H3 Max | 1 | 1195 | 1186-1204 | 5894 | 5 (1130) |
| Vidu Q4 Preview | 3 | 1179 | 1169-1189 | 5543 | none |
| Gemini Omni Flash | 4 | 1178 | 1171-1185 | 12 327 | 8 (version 1.1, 1113) |
| Seedance 2.0 720p | 5 | 1176 | 1169-1183 | 15 721 | 7 (1116) |
| HiDream-O1-Video-1.0 | 6 | 1175 | 1165-1185 | 3231 | none |
Data: Artificial Analysis, image-to-video and text-to-video, retrieved October 11, 2026, intervals rounded. Holding 2nd place in I2V is MiniMax H3 (1181), which I cover below regarding open weights.
Scale starts at 1000 points (bar = (Elo - 1000) / 2 percent). How to interpret this:
- Vidu Q4, Gemini, Seedance 2.0, and HiDream are tied. Their confidence intervals overlap, with merely 4 points separating 3rd and 6th place.
- H3 Max is clearly ahead. Its lower bound (1186) sits above the upper bounds of HiDream (1185) and Seedance (1183). With Vidu Q4, the intervals barely touch.
- HiDream has the fewest matchups (3231), making its ranking the least certain in either direction.
Cost Per Minute of Video
| Model | USD per Minute (AA) | 10s Clip (Calculated) | Official Price List |
|---|---|---|---|
| MiniMax H3 Max | 4.80 | $0.80 | yes (via providers) |
| Vidu Q4 Preview | 5.04 | $0.84 | yes (Vidu credits) |
| HiDream-O1-Video | 5.80 | $0.97 | not found |
| Gemini Omni Flash | 6.00 | $1.00 | not verified |
| Seedance 2.0 720p | 9.07 | $1.51 | yes (tokens) |
Scale: 100% = $9.07. Prices reflect the Artificial Analysis benchmark table (each model in its default setup). The 10-second clip figure is my own calculation. Keep in mind that for Vidu Q4, AA reported $7.20 on October 8, adjusting it to $5.04 on October 11. For HiDream, $5.80 is the only figure available; HiDream does not publish an official price list. For more on billing structures (seconds, credits, tokens), see my breakdown on how much a second of AI video costs.
Side-by-Side Specifications
| Feature | HiDream-O1-Video | MiniMax H3 Max | Vidu Q4 Preview | Seedance 2.0 |
|---|---|---|---|---|
| Text-to-video | not in API (announced) | yes | no | yes |
| API inputs | 1 image + prompt | text, first and last frame | 1 image or up to 15 references + 3 audio voices | text, images, video, audio |
| Duration | model-chosen or 10s (announced: 5-20s) | 5-15s | 3-16s | 4-15s |
| Resolution | no parameter (announced: 1080p) | 480p, 768p, 1080p | 540p-4K | 480p-4K |
| Audio | announced, undocumented in API | yes (fal.ai) | yes, default | yes |
| Aggregators | none | fal.ai, OpenRouter, reAPI | fal.ai, WaveSpeedAI, ComfyUI | BytePlus, OpenRouter, fal.ai, reAPI |
| Open weights | no (only O1-Image is open) | no (weights available for H3, not Max) | no | no |
Sources: HiDream - HiHarness documentation (October 11, 2026); remaining models - data from my guides covering H3 Max, Vidu Q4, and Seedance (as of October 8, 2026); OpenRouter - video model list as of October 11, 2026 (includes minimax/hailuo-3-max and bytedance/seedance-2.0, but lacks Vidu and HiDream). Gemini Omni Flash is omitted from this table; see the Gemini Omni Flash guide for its technical specs.
Open Weights: Only H3, With Major Caveats
In the AA leaderboard, MiniMax H3 (not H3 Max) is the only model in this tier offering a link to open weights. On its model card (retrieved October 11, 2026), only the H3-Base component is available (33B parameters) under the MiniMax H3 Community License, which requires an access form. The reference deployment requires 4 GPUs, while the 2K upscaler still relies on their API. This is not a model you can run on a typical home desktop.
HiDream does not release weights for O1-Video. The MIT-licensed release is HiDream-O1-Image, an image generation model that works well for creating starting frames. You can read more about self-hosted alternatives in the local AI guide.
Global Availability
- MiniMax H3 Max and Seedance 2.0 - easiest to access via OpenRouter or fal.ai: a single account, standard card billing, and multi-model routing.
- Vidu Q4 - vidu.com, the Vidu API, and fal.ai. It is not listed on OpenRouter.
- HiDream-O1-Video - strictly available through vivago.ai and the HiHarness API. It remains unclear whether vivago has fully deployed the model in R1 Studio: a post from September 30 states “now available,” while an announcement from October 9 refers to an “upcoming update.” I analyzed this in HiDream’s API is limited to image-to-video.
Which Model Should You Choose?
- MiniMax H3 Max - when you want the highest leaderboard position at the lowest price per minute, along with text-to-video capability. The most practical choice for most projects.
- Vidu Q4 Preview - when generating recurring characters and matching voiceovers, or when you require 4K output and up to 15 reference inputs. The trade-off: an initial image is always required.
- Seedance 2.0 - when you prefer unified billing through an aggregator like OpenRouter or fal.ai and need every possible input modality. The trade-off: the highest per-minute cost in this tier.
- HiDream-O1-Video - when animating pre-rendered scenes governed by physics (bounces, collisions, surface dynamics) and managing a separate platform account is not an issue. The trade-off: limited API options, unlisted pricing, and no manual control over resolution or audio.
- Gemini Omni Flash - when your workflow is tied directly to the Google ecosystem; its image-to-video performance benchmarks on par with HiDream.
If your primary requirement is text-to-video, consider looking into Wan 3.0, which holds 1st place in T2V. For a comprehensive market breakdown, see my guide to AI video generators.
Frequently Asked Questions
Is HiDream-O1-Video better than Vidu Q4?
The leaderboard cannot definitively separate them: scores of 1175 vs. 1179 with overlapping confidence intervals indicate a statistical tie. Vidu Q4 is cheaper ($5.04 vs. $5.80 per minute on AA) and provides significantly more configuration options.
Which model is the cheapest?
MiniMax H3 Max - priced at $4.80 per minute on Artificial Analysis (October 11, 2026). HiDream ranks third in cost.
Which one offers open weights?
Only MiniMax H3 (the base model, not H3 Max), and only under an approved community license. HiDream has open-sourced only its O1-Image generation model.
Can I run these via OpenRouter?
You can use MiniMax H3 Max and Seedance 2.0 (as of the October 11, 2026 catalog). HiDream and Vidu Q4 are currently unavailable there.
Google · Your sourcesQuantslant higher in your Google - one clickAdd to preferred sources →Sources and Verification Date
- Artificial Analysis, Image-to-Video and Text-to-Video leaderboards (retrieved October 11, 2026)
- HiHarness Documentation (October 11, 2026)
- MiniMax-H3 on Hugging Face (October 11, 2026)
- OrcaRouter: HiDream-O1-Video vs MiniMax H3 (approx. October 10, 2026)
- OpenRouter Video Models Endpoint,
/api/v1/videos/models(October 11, 2026)
