MiniMax H3-Image
MiniMax·
Overview
Open-weight image model from MiniMax: a dense 33.1B single-stream diffusion transformer (50 layers) built on the H3 design with the audio branch removed, distilled to 6 sampling steps without CFG. Text-to-image and editing with up to 16 reference images, up to about 16 MP (4096x4096 at 1:1). It uses the unchanged MiniMax-H3 text encoder, which carries the full pretrained weights of Qwen3-VL-32B; the repository totals about 204 GB. MiniMax states no VRAM requirement. MiniMax-H3-Image Community License: it does not apply in the European Union, the United Kingdom, South Korea or the United States, including use of outputs there, and licensees above US$1M annualized revenue or offering hosted services need separate written authorization. MiniMax publishes no benchmark figures.
Capabilities and innovations
Benchmarks
No benchmark values are recorded for this model.
- GPQA Diamond: not reported by the vendor
- Humanity’s Last Exam (no tools): not reported by the vendor
- SWE-bench Verified: not reported by the vendor
- SWE-bench Pro: not reported by the vendor
- MMLU: not reported by the vendor
- MMLU-Pro: not reported by the vendor
- Terminal-Bench 2.x: not reported by the vendor
- MMMU-Pro: not reported by the vendor
Architecture and hardware
- Parameters
- 33.1B total (Dense)
- Estimated VRAM at Q4
- ~20 GB, Edge / Consumer class
Links
More from MiniMax
Data curated by AI Model Timeline. See the methodology for admission criteria, benchmark eras and source priority.