MiniMax M3 vs H3

MiniMax H3 is a video generation model. MiniMax M3 is a language, coding, and agent model. They are different products with different outputs — this page explains which is which.

Quick answer

MiniMax H3 generates video; MiniMax M3 handles language, coding, and agent tasks. MiniMax3.org provides access to MiniMax H3 video workflows only. The two models are not interchangeable.

  • H3 → video clips (4–15 s, 768p/2K), via text, image, and reference inputs.
  • M3 → text, code, and tool/agent actions; not offered on MiniMax3.org.
  • Last verified: 2026-08-11

Side-by-side comparison

MiniMax H3 vs MiniMax M3 comparison
FieldMiniMax H3MiniMax M3
Model typeVideo generationNative multimodal language, coding, and agent model
Primary inputText, image, video, audio referencesText, image, and video context
Primary outputVideo clips (MP4)Text, code, tool / agent actions
Resolutions768p, 2K
Duration4–15 seconds
Context windowNot applicableUp to 1M tokens
Available on MiniMax3.orgYesNo

Last verified: 2026-08-11. Sources: MiniMax M3 official repository and MiniMax H3 official documentation. Provider documentation can change.

MiniMax H3 — the video model

MiniMax H3 is a multimodal video generation model. It accepts text prompts, image references (first and/or last frame), video references, and audio references, and returns short video clips. On MiniMax3.org it is available through text-to-video, image-to-video, and reference workflows at 768p or 2K for 4–15 seconds. See the MiniMax H3 guide for details.

MiniMax M3 — the language and coding model

MiniMax M3 is MiniMax's native multimodal language, coding, and agent model with a documented 1M-token context window. It can reason over text, image, and video context, but its primary outputs are language, code, and tool or agent actions rather than generated video. It is not offered on MiniMax3.org.

Is "MiniMax 3" the same as H3 or M3?

"MiniMax 3" is an informal search spelling that can refer to either model. On MiniMax3.org, the term primarily refers to the video-generation intent and is handled through MiniMax H3. Where the language/coding model is meant, the official name MiniMax M3 is used.