Calculator
How much of your video will the model see?
Pick a model and a video length. For every provider of that model, this shows how much of the video fits in one request, sent as frames or as a video file, from the limits we measured.
frames the model sees in one request, the better of the two ways to send it
How this is worked out
- As frames: the most 768×432 frames the provider accepted in one request, spread evenly over your video. Larger frames fit fewer. Where only the image-count test ran, the image cap is shown for reference but not ranked: real frames often hit other limits first.
- Tokens are the provider's measured prompt tokens per 768×432 frame, times the frames.
- As a video file: whether your file fits the provider's size limit, and how many frames it then feeds the model. That rate comes from a 5-minute reference clip; where the provider caps frames per video (⁎), a longer video gets the same number spread thinner. For other providers, videos longer than 5 minutes are extrapolated, and may hit duration limits we haven't measured. A file larger than any we tested isn't ranked.
- All of it is for a single request. See the methodology for how each limit is measured.