Gen-4.5
Gen-4.5 took #1 on Artificial Analysis text-to-video (1,247 Elo) at launch, ahead of Veo 3, Kling 2.5 and Sora 2 Pro.
Improves prompt adherence, physical accuracy and expressive characters at similar pricing to Gen-4, and was trained and served on NVIDIA GPUs (Hopper, Blackwell). Runway admits remaining failures in causal reasoning, object permanence and 'success bias'. The lead was short-lived, and by May 2026 it had dropped out of the top 10.
- Date
- Monday, 1 December 2025
- Lab
- Runway
- Kind
- model
- Access
- app only
Figures
| Measure | Value | Measured by |
|---|---|---|
| Artificial Analysis T2V Elo | 1,247 vs Veo 3 1,226, Kling 2.5 1,225, Sora 2 Pro 1,206, at launch | third-party leaderboard, cited by company |
Ranking claim is on Runway's own page; the 'dropped out of top 10 by May 2026' comes from a secondary roundup and was not verified.
Sources
This record was checked against its sources on 6 October 2026. How we check