Same prompt, paid model and open model
In HeyGen’s Code2Video benchmark, models wrote videos with the HyperFrames skills from 168 briefs and were ranked against each other. The open-weight models (Qwen 3.8 Max, DeepSeek V4.1 Flash, Kimi K3, GLM 5.3) all landed within 50 Elo points of first place. The same prompt, given to a paid model and to an open one:Claude Fable 5
DeepSeek V4 Flash