# Pika releases H3 Flex, a six-step open video and audio model post-trained on MiniMax's H3

The 33-billion-parameter model is distilled from 50 sampling steps to six and ships as two LoRA adapters on the Chinese base model, and it ranks seventh on DesignArena's text-to-video board.

- Published: 2026-11-01T06:20:01.798Z
- Canonical: https://polylog.news/ai/2026-11-01/pika-releases-h3-flex-a-six-step-open-video-and-audio-model
- Publisher: Polylog (AI desk)
- Section: tech
- Sources: [Polylog editors](https://polylog.news), [Hugging Face](https://huggingface.co/Pika-Labs/Pika-H3-Flex), [AI Weekly](https://aiweekly.co/alerts/pika-labs-opens-33b-h3-flex-audio-video-model-on-hugging-face), [ComfyUI Wiki](https://comfyui-wiki.com/en/news/2026-10-29-pika-h3-flex), [ComfyUI Docs](https://docs.comfy.org/tutorials/video/pika/pika-h3-flex), [explainx](https://explainx.ai/blog/pika-h3-flex-open-weight-video-audio-model-minimax-h3-six-steps-designarena-7-comfyui-license-2026)

Pika Labs released H3 Flex, a 33-billion-parameter video model with synchronized audio. It is post-trained from MiniMax-H3, the open-weight video model from the Chinese lab MiniMax. According to the Russian-language channel AI ML Big Data,…

This story is for subscribers. Read it in full at https://polylog.news/ai/2026-11-01/pika-releases-h3-flex-a-six-step-open-video-and-audio-model (subscription information: https://polylog.news/pricing).