Head-to-Head · Video AI
Wan 2.2 vs LTX-2.3 — Open Video Models
The two strongest open-weights video models you can self-host today.
Verdict
LTX-2.3 Fast wins on speed, resolution and clip length; Wan 2.2 counters with the MoE architecture community momentum.
Side-by-Side Comparison
| # | Model | Arena Elo | Gen Latency | Cost / 1K Clips | Max Output | License |
|---|---|---|---|---|---|---|
| 14 | Wan 2.2 A14B Alibaba | 1174 | 55s | $70 | 720p / 5s | Open Weights |
| 13 | LTX-2.3 Fast Lightricks | 1187 | 20s | $60 | 1080p / 20s | Open Weights |
Model-by-Model
Wan 2.2 A14B — Alibaba
LTX-2.3 Fast (1187 Elo) generates in ~20s at just $0.06/clip API — and as open weights it can run entirely on your own hardware at 1080p/20s.
LTX-2.3 Fast — Lightricks
Wan 2.2 A14B (1174 Elo) is Alibaba’s MoE video model — open weights, $0.07/clip, but limited to 720p/5s output.
FAQ
What is the best open-source video model?
LTX-2.3 Pro/Fast currently lead open-weights video at 1225/1187 Elo, ahead of Wan 2.2 (1174).
Can I self-host these models?
Yes — both ship open weights. LTX-2.3 needs a high-VRAM GPU (or use the API at $0.06/clip); Wan 2.2’s MoE design is heavier to serve.