HappyHorse 1.1 vs Wan 3.0 vs Seedance 2.5 vs MiniMax H3 Max
Four AI video models tested across identical scenes: Artificial Analysis Elo, per-minute pricing, 5s clip costs, benchmark scores, and key features compared.
👁 129 reads
Across the same four benchmark scenes in my Video Arena, HappyHorse 1.1 averaged 6.1/10, Wan 3.0 reached 9.1, MiniMax H3 Max hit 8.25, and Seedance 2.5 scored 6.0. The Artificial Analysis leaderboard reflects a similar gap: HappyHorse 1.1 sits in 12th place for text-to-video, while its three rivals all land inside the top five. Below is a detailed breakdown of benchmark rankings, costs, hands-on scores, and core features. For an in-depth dive into the model itself, see my HappyHorse 1.1 guide.
Verdict at a Glance (October 11, 2026)
- Best overall quality: Wan 3.0 - 1st on the AA text-to-video leaderboard and 9.1/10 in my tests.
- Best image-to-video and multilingual speech: MiniMax H3 Max - 1st in image-to-video, $4.80 per minute, 0 WER on Polish speech tests.
- Longest clips and editing flexibility: Seedance 2.5 - up to 30s clips, references, video extension tools, but the highest price tag.
- HappyHorse 1.1: solid camera work and consistent characters, but trails behind in text rendering, physical dynamics, and non-English speech.
Artificial Analysis Leaderboard Rankings
| Model | Text-to-Video: Rank, Elo | 95% Confidence Interval | Image-to-Video: Rank, Elo | 95% Confidence Interval |
|---|---|---|---|---|
| Wan 3.0 | 1st, 1156 | 1147-1165 | 7th, 1164 | 1156-1172 |
| Seedance 2.5 | 3rd, 1143 | 1134-1152 | not listed (Seedance 2.0 720p: 5th, 1176) | 1169-1183 |
| MiniMax H3 Max | 5th, 1130 | 1120-1140 | 1st, 1195 | 1186-1204 |
| HappyHorse 1.1 | 12th, 1042 | 1033-1051 | 8th, 1104 | 1097-1111 |
Data captured on October 11, 2026, from the Artificial Analysis text-to-video and image-to-video leaderboards. In text-to-video, the confidence intervals for Wan 3.0 and Seedance 2.5 overlap, making it a statistical tie. HappyHorse 1.1 trails them by over 100 Elo points and lags almost 90 points behind H3 Max. In image-to-video, HappyHorse sits 60 points behind Wan 3.0 and over 90 points below H3 Max.
Scale starts at 1,000 points; full bar corresponds to 1,200.
Cost per Minute and 5-Second Clip Pricing
| Model | USD/min T2V (AA) | USD/min I2V (AA) | My 5s Clip Cost | OpenRouter 5s Rate (Oct 11) |
|---|---|---|---|---|
| MiniMax H3 Max | $4.80 | $4.80 | $0.25 (480p, Oct 8) | $0.25 (480p) / $0.40 (768p) |
| HappyHorse 1.1 | $10.80 | $9.90 | $0.494 (720p, Oct 11) | $0.494 (720p) / $0.64 (1080p) |
| Wan 3.0 | $12.00 | $12.00 | $0.21 (480p, Oct 8) | $0.25 (480p) / $0.50 (720p) |
| Seedance 2.5 | $34.12 | not listed | $0.52 (480p, Oct 8) | token-based billing |
Per-minute figures originate from the Artificial Analysis benchmark tables (AA default configuration). OpenRouter pricing reflects catalog listings as of October 11, while individual clip figures come directly from the usage.cost parameter returned in API responses. At 720p on OpenRouter, HappyHorse 1.1 and Wan 3.0 cost virtually the same ($0.494 vs $0.50 per 5 seconds). H3 Max is the most budget-friendly, while Seedance 2.5 is the priciest. For a broader overview, check my breakdown on how much AI video generation costs per second.
Hands-on Arena Test Scores
The Video Arena subjects every model to five identical test scenarios using an identical evaluation framework (evaluating four individual frames alongside the complete clip, scored 0-10). Wan 3.0, H3 Max, and Seedance 2.5 were benchmarked on October 8, while HappyHorse 1.1 was tested on October 11. One operational difference: the other models ran at 480p, whereas HappyHorse 1.1 was evaluated at 720p because OpenRouter lacks a 480p tier for it. That extra resolution, however, did not rescue its performance on scenes where it stumbled.
| Scene | Wan 3.0 | MiniMax H3 Max | Seedance 2.5 | HappyHorse 1.1 |
|---|---|---|---|---|
| Motorcycle camera orbit | 8.5 | 8 | 6 | 8 |
| Chalkboard text | 10 | 10 | 5 | 3 |
| Pouring water | 9 | 7.5 | 7 | 5 |
| Two-shot character cut | 9 | 7.5 | 6 | 8.5 |
| Average across 4 scenes | 9.1 | 8.25 | 6.0 | 6.1 |
| Polish speech WER (lower is better) | 0.167 | 0 | 0.583 | 0.5 |
Full bar indicates 10 points. WER scores for Wan 3.0, H3 Max, and Seedance 2.5 come from our Polish speech evaluation round on September 28 (using identical scenes and prompts; we tested speech generation specifically in Polish). H3 Max was driven via fal.ai for that speech test because its OpenRouter endpoint does not support audio generation.
Where HappyHorse 1.1 Holds Its Ground
On the 180-degree camera orbit prompt, HappyHorse scored an 8, matching H3 Max and trailing Wan 3.0 by just half a point. In the multi-shot character scene (8.5), it outpaced both H3 Max and Seedance 2.5. The character’s face, glasses, and coat stayed consistent through the scene cut, with only slight hair length drift.
Where HappyHorse 1.1 Falls Short
In-frame lettering showed the starkest contrast: Wan 3.0 and H3 Max spelled both lines flawlessly, correctly preserving complex Polish diacritics. HappyHorse 1.1 dropped letters in the word for video (translated from the Polish original) and corrupted the Polish phrase into [corrupted lettering in the original Polish test]. Seedance 2.5 struggled too, hallucinating unrelated German and Czech accents. During the liquid simulation test, HappyHorse formed a puddle beneath the glass before the container had even filled.
Feature Comparison
| Feature | HappyHorse 1.1 | Wan 3.0 | Seedance 2.5 | MiniMax H3 Max |
|---|---|---|---|---|
| Release date | June 22, 2026 | August 19, 2026 | July 31, 2026 | August 26, 2026 |
| Clip length | 3-15s | 2-30s | 4-30s | 5-15s |
| Resolution | 480p-1080p (OpenRouter: 720p, 1080p) | 480p-1080p | 480p, 720p (OpenRouter) | 480p, 768p (OpenRouter) |
| Image-to-video | First frame | First frame | First and last frame | First and last frame |
| Reference inputs | 1-9 images | Yes (per OpenRouter specs) | Multimodal, plus editing and video extension | Not via OpenRouter |
| Audio generation | Yes | Yes | Yes | No via OpenRouter; yes via fal.ai |
| Open weights | No | No | No | No (standard MiniMax H3 weights are open) |
Release dates per Artificial Analysis; feature specs sourced from Alibaba Cloud documentation (HappyHorse) and the OpenRouter model registry (all others) as of October 11, 2026. None of these four flagship models can be self-hosted locally. For self-hosted alternatives, see our guide to local open-weight AI models.
Which Model Should You Choose?
- Wan 3.0 - Choose this when pure prompt fidelity, clean typographic rendering, and natural physics matter most. It took first place in my hands-on ratings and on the text-to-video leaderboard, while matching HappyHorse 1.1 pricing at 720p on OpenRouter. It also supports clips up to 30 seconds. Read more: Wan 3.0 review.
- MiniMax H3 Max - Your best bet if you work primarily from input images, need reliable multilingual lip sync, or want the lowest per-minute costs. Ranked #1 in image-to-video and cheapest per minute. Note that its OpenRouter endpoint does not currently deliver audio. Read more: MiniMax H3 Max review.
- Seedance 2.5 - Opt for this if your pipeline demands 30-second sequences, multi-image conditioning, or iterative in-video expansion. While it ties Wan on the AA leaderboard, it struggled in my specific test scenes and has a significantly higher price point. Read more: Seedance review.
- HappyHorse 1.1 - Well suited for concise clips focused on dynamic camera movements, consistent character framing across cuts, or basic English dialogue, especially when taking advantage of affordable promotional pricing in the app (Standard tier: $1 for 7 days). Avoid it if your project relies on in-frame text or non-English speech.
Comprehensive comparisons covering other contenders like Kling, Veo, and Runway are available in the Video Arena.
Frequently Asked Questions
Is HappyHorse 1.1 better than Wan 3.0?
No. Wan 3.0 ranks higher across both Artificial Analysis leaderboards and achieved an average score of 9.1 vs 6.1 in my direct testing. While both originate under Alibaba’s corporate umbrella, they are engineered by separate teams.
Which of the four models handles non-English speech best?
MiniMax H3 Max leads decisively, achieving a 0 Word Error Rate (WER) in my Polish tongue-twister benchmark. Wan 3.0 missed 2 words out of 12, HappyHorse 1.1 failed on 6, and Seedance 2.5 dropped 7.
Which model is the most affordable?
MiniMax H3 Max is the most cost-effective option, listed at $4.80 per minute on the AA leaderboard and $0.25 per 5-second clip at 480p on OpenRouter.
Sources and Verification Dates
- Artificial Analysis Text-to-Video and Image-to-Video Leaderboards (Checked October 11, 2026)
- OpenRouter Video Model Directory (Checked October 11, 2026)
- Alibaba Cloud Model Studio Documentation (Checked October 11, 2026)
- Hands-on evaluations: Video Arena (October 8, 2026; Polish WER round from September 28, 2026) and HappyHorse 1.1 evaluation (October 11, 2026)
