Published • loading... • Updated
MiniMax M1 Model Claims Chinese LLM Crown From DeepSeek
CHINA, JUN 17 – MiniMax's M1 offers a 1 million token context window with 70% less compute needed than DeepSeek R1, aiming to lead China’s AI market with open-source innovation and efficiency.
- Shanghai-Based AI firm MiniMax released its open-source M1 reasoning model with a one million-token context window on June 16, 2025.
- MiniMax developed M1 to challenge DeepSeek's R1 model and major overseas AI systems amid intense competition and ongoing IPO discussions in Hong Kong.
- M1 uses the Lightning Attention mechanism and a novel reinforcement learning approach to achieve up to 70% lower compute costs than DeepSeek R1 for deep reasoning.
- Training the reinforcement learning model required three weeks on a cluster of 512 Nvidia H800 GPUs, incurring a cost of $537,400—substantially less than DeepSeek R1's $5.6 million—demonstrating notable gains in computational efficiency.
- MiniMax's move to open-source M1 aims to reduce AI adoption barriers, potentially disrupting cloud providers and Western proprietary models while advancing China's AI leadership.
Insights by Ground AI
11 Articles
11 Articles
: China's 'little dragons' pose big challenge to US AI firms
China’s AI Challenger MiniMax: Open Models, Multi-modal Products, and IPO Drive
Pioneering Open Models: From Abab to M1Shanghai-based AI startup MiniMax has quickly emerged as a model-driven company, pushing the boundaries of large language models (LLMs). In January 2024, it launched Abab6, China’s first Mixture-of-Experts (MoE) LLM , and followed up in April with Abab 6.5, a trillion-parameter model supporting 200,000-token context lengths . The Abab 6.5 series (including an optimized 6.5s variant) was shown to approach th…
Coverage Details
Total News Sources11
Leaning Left0Leaning Right0Center3Last UpdatedBias Distribution100% Center
Bias Distribution
- 100% of the sources are Center
100% Center
C 100%
Factuality
To view factuality data please Upgrade to Premium
