MiniMax open-sources H3-Image: 4K image generation with CAD-level reasoning

量子位 · wechat · 2026-11-12

After open-sourcing its H3 video model, MiniMax has released H3-Image, an open-weights flagship image model that impressed reviewers in hands-on tests.

Realistic skin detail (pores, freckles, vellus hair) and deliberate imperfections largely eliminate the usual "AI look."
Strong spatial reasoning: generates standard orthographic CAD views from product photos and mechanically accurate exploded diagrams with labeled components.
Handles dense academic typesetting (calculus exam papers, no garbled symbols) and even produces correct solutions.
Specs: native 16MP (4096×4096) output in 10-20 seconds, up to 16 reference images for character consistency and storyboard continuity.
Open weights, with API pricing at 50-70% of mainstream competitors and a first-week 50% discount on 16MP requests.

The author argues multimodal generation may be approaching its own "Coding moment" as quality, speed, openness and price align.

Original post →

MiniMax open-sources H3-Image: 4K image generation with CAD-level reasoning

More from Multimodal