88Z AI, 1 June — MiniMax has released the open-source MiniMax M3, the first domestic model combining advanced coding, million-token context, and native multimodal capabilities. Utilizing a new MSA sparse attention architecture, it achieves strong performance in international benchmarks, surpassing some leading overseas models.
At a glance
- First domestic open-source model combining coding, million-token context, and multimodality
- Uses MSA sparse attention architecture, outperforming GPT-5.5 in benchmarks
The update
MiniMax has officially launched the new generation open-source AI model, MiniMax M3. This model is the first in China to simultaneously feature advanced coding capabilities, support a long context window of one million tokens, and possess native multimodal abilities. Technically, M3 adopts a novel MSA sparse attention mechanism, an innovation that significantly enhances processing efficiency. In the authoritative international benchmark SWE-Bench Pro, MiniMax M3 demonstrated strong competitiveness. Its key metrics not only surpassed well-known overseas models such as GPT-5.5 but also approached the leading levels of Opus 4.7. This release marks a significant breakthrough for domestic open-source models in handling complex tasks and long-text understanding, providing developers with a high-performance technical option.
Sources & editorial note
This article is an AI-assisted synthesis by 88Z AI based on the public sources below. Product claims and availability are attributed to their publishers. It does not represent independent testing or first-hand reporting.
- MiniMax 稀宇科技Source published
Source dates above refer to the original announcements; the date under the headline is this article’s publication on 88Z AI.