# MiniMax H3 ecosystem gains 8-step reference-conditioned video+audio, 4-step distillation and new serving integrations

MiniMax · 2026-10-25 · Capability update

> MiniMax reports community contributions to the H3 video model: 8-step reference-conditioned video with audio, 4-step distillation, sparse attention, reference-token compression, and serving/workflow integrations across GPU servers, TPUs and Apple Silicon, with 11 additions to the H3 ecosystem index.

This update bears on 3 kinds of work in 2 markets. Highest level accepted: L1 Assisted.

## The work it bears on

- [Develop algorithms and study computational properties](https://surviagi.com/work/science-research-computing-engineering-algorithm-research) (Computing and engineering research): L1 Assisted · From the publisher only
- [Build research prototypes and run comparative experiments](https://surviagi.com/work/science-research-computing-engineering-engineering-experiments) (Computing and engineering research): L1 Assisted · From the publisher only
- [Integrate external systems and data interfaces](https://surviagi.com/work/software-application-code-integration-code) (Application interfaces and business development): L1 Assisted · From the publisher only

## Published at

- [Source 1](https://x.com/MiniMax_AI/status/2114504329499660419)

---

Page: https://surviagi.com/updates/ev-d541888c0f691f17eb4740c605af8fd113d5e318

Data generated 2026-10-26T00:14:10.392773+00:00; method activity-level-1; ontology 2.4.0.
Machine-proposed: no person has reviewed these readings, and the method is not calibrated.
Cite the page address together with the generation time.