pika drops h3 flex, a fast video model with open weights

what it is

pika released h3 flex, its post-trained version of the minimax h3 video model.

pika says it trained it to generate in six steps instead of 50, then refined it for quality and control.

what it can do

per pika:

• up to 12 references in one generation
• the character from one image, the world from another, motion from a video
• five separate audio stems
• output up to 4k
• a 15-second clip with synced audio at 768p in as little as seven seconds of compute

pika says flex ranked #1 among open-weights models and #7 overall on designarena's text-to-video leaderboard at the time of writing.

the open part

pika is releasing open weights. you can build on the model itself instead of only renting it through an app.

check the licence of the underlying model before you use it commercially. more on that in our minimax h3 piece.

why it matters for a creator business

character consistency plus motion from a reference clip is the core of most short-form video work.

if a model does that well and you can run it yourself, your cost per clip changes.