Generative-media models bill per output, not per token. These are the live fal.ai hosted-inference prices across fal's full catalog — 1,266 models, 1,022 with a flat per-output price (per image, megapixel, or video-second) — spanning image, video, audio and 3D. GPU/compute-billed models are kept separate (toggle in the table) since their rate isn't a per-output price. The token-based Provider Pricing Index covers LLMs; this covers media.
Prices are fal.ai's hosted-inference rates, pulled from the fal.ai pricing API. Each model bills per its own unit (image, megapixel, video-second, characters…), so the numbers aren't directly comparable across rows — a per-megapixel price scales with resolution, a per-second price with length. The table opens on popular models; switch to the full catalog with the dropdown. Models billed by GPU compute-second, step, or token are a pay-as-you-compute rate (not a per-output price), grouped under “GPU / compute-billed.” Looking for quality rankings instead? See the generative media leaderboard. Updated May 31, 2026.
