Release2026-10-29

Pika released Pika H3 Flex, a post-trained MiniMax H3 model. It generates video faster than real time. It accepts up to 12 reference files for images, video, and audio. It exports five separate audio stems. It finishes at up to 4K resolution. Open weights are available for customization. It ranks first among open-weights models on DesignArena’s leaderboard. It will launch in the Pika creation product on November 2. Pika-hosted features include upscaling and audio-stem separation.

What changed

MiniMax H3 required 50 denoising steps and did not support multi-reference or multi-stem audio generation.

What it unlocks

Generating video from up to 12 references, exporting five separate audio stems, and finishing at up to 4K resolution.

  • up to 12 reference files
  • 5 separate audio stems
  • 7 seconds for 15-second 768p clip
  • #1 among open-weights models on DesignArena

What you need to act on it

  • access to Pika API Club or Pika creation product
  • own hardware for open weights usage

Send this to someone who needs it

Shares the story and its sources. Nothing about you.

What does this mean for your job?

This is the story as everyone gets it. Once a week we send you the version written for your role — what changed, why it matters for the work you actually do, and one thing to try. Free while we tune it.