CineSubBench Evaluates Long-Context Multilingual Film Understanding
September 30, 2026
CineSubBench provides a benchmark for assessing long-context narrative and cultural reasoning using 8.13M timestamped subtitle entries across 1,012 films. It tests models on seven tasks including character reconstruction and causal progression using six different languages.
HOW THIS AFFECTS YOU
●
researcherYou can use this to evaluate how well your models handle long-form, multilingual narrative coherence.