Video model leaderboard rank: how to turn Elo into an API choice
Hedra's board puts Seedance 2.0 at Elo 1,225 (#2) and HappyHorse 1.1 at 1,149 (#4). What a rank tells an API buyer and what to test instead.

A leaderboard rank tells you how viewers voted on side-by-side clips, not which model will do your job. On Hedra's roundup, read 2026-10-02, Seedance 2.0 sits at Elo 1,225 (#2) and HappyHorse 1.1 at Elo 1,149 (#4). Use that as a shortlist signal, then pick an API model by clip length, resolution, inputs and price, and test it on your own prompts.
The rank figures come from Hedra's best AI video models page, read 2026-10-02. Sume's side comes from the Video Router docs and Video generation docs.
What do Elo 1,225 and Elo 1,149 actually say?
On a standard Elo scale, a gap of 76 points means the higher-rated model is expected to win about 61 percent of head-to-head votes, not every vote. Hedra's text as read does not describe its scale, so treat that figure as a rule of thumb, not a published claim.
A vote gap also says nothing about your constraints. A model can rank high and still cap at 15 s, lack audio references, or not be on the API you use.
| Model | Hedra rank | Hedra Elo | On Sume's catalog? |
|---|---|---|---|
| Seedance 2.0 | #2 | 1,225 | Yes, as seedance-2 |
| HappyHorse 1.1 | #4 | 1,149 | No catalog id |
Which constraints should decide before rank does?
Start with what the clip must be. Sume's catalog gives per-model clip length and resolution: seedance-2.5 takes 4-30 s, wan-3.0 takes 2-30 s, minimax-h3 takes 5-15 s, and seedance-2 takes 4-15 s with 1080p available. If a job needs a 25-second clip, every model capped at 15 s is out before anyone votes.
Next check inputs: first and last frames, image, video and audio references. Sume's docs say audio and video references are honored by the Seedance 2.x models, Wan 3.0, MiniMax H3 and MiniMax H3 Max, while Gemini Omni Flash 1.1 takes video references but not audio.
How do I test candidates cheaply?
Run one prompt through two or three ids at 480p and 5 to 8 seconds, with a different Idempotency-Key per request. Sume reserves the estimated cost on submit and captures it on completion, so a draft pass is a small spend. The post same prompt on three models has the code.
Then rank by your own rubric: does the product stay on model, is the text legible, does motion match the brief. Write the rubric before you look at the clips.
Sources
Related posts
More in Models
- Which AI video model gives 1080p on Sume, and which stop at 768p?
Sume lists 1080p for Seedance, Wan 3.0, Kling 3.0 and Auto; MiniMax H3 is native 480p or 768p in the panel. Resolution table by model, with the API check.
- Voxtral TTS: open weights under CC BY-NC versus the paid API
Mistral's Voxtral TTS has open weights under CC BY-NC 4.0 and a paid API. What the licence means for a product, and what Sume offers instead.
- Wan 2.2 A14B explained: 27B total, 14B active, 80GB GPU
Wan 2.2's A14B models use two experts, one for noisy early steps and one for late detail. What that means for GPU memory, and when a hosted route is simpler.
- Wan 2.2 weights: Apache 2.0, 5-second 480p/720p; Sume hosts wan-3.0
Wan 2.2 T2V-A14B is Apache 2.0 with open weights and makes 5-second 480P or 720P clips. Sume does not host Wan 2.2; its Wan row is wan-3.0, 2 to 30 seconds.
Written by Sume