MiniMax open-sources native multimodal flagship model M3 with 428B total parameters
Original title:我们开源了 MiniMax M3
AI summary
MiniMax open-sourced the M3 model weights last Friday and published a paper on its MSA sparse attention technique. M3 has 428B total parameters and 23B activated parameters, making it the first open-source model trained with multimodal mixture from Step 0. Two weeks after release, it ranked first among global open-source models on the Artificial Analysis composite intelligence index. Output speed has increased from about 30 TPS to about 80 TPS, and the company says it will further increase speed by 30-40%.
Key Takeaways
- MiniMax open-sourced the M3 model weights last Friday and published a paper on its MSA sparse…
- M3 has 428B total parameters and 23B activated parameters,making it the first open-source model…
- Two weeks after release,it ranked first among global open-source models on the Artificial Analysis…
- Output speed has increased from about 30 TPS to about 80 TPS,and the company says it will further…
Quick Facts
- Published
- Updated
- Source
- 公众号:MiniMax(稀宇科技)
- Category
- Tutorials & Tips
- AI score
- 72
MSA 注意力机制让长上下文计算成本大幅下降,对于在长文档处理或复杂推理场景下需要控制成本的团队,M3 的开源提供了一种新的选型依据。
Related entities
Source & copyright
This article is based on curated titles and AI summaries from the AIHOT public API, with attribution and links back. Original copyright remains with the original publishers and authors; this site does not mirror full texts.
How to Cite
When citing this article, keep the title, date, and permalink, and credit Orange News as the source.
Orange News: MiniMax open-sources native multimodal flagship model M3 with 428B total parameters. June 15, 2026. https://new.nat6.net/en/news/cmsdu3hys0160rofzq35jx504/