Shanghai-based AI company MiniMax has launched its latest large language model, MiniMax-M1, which offers a context length of one million tokens.

This capacity is eight times larger than that of DeepSeek’s R1 model, allowing for the simultaneous processing of more information.

MiniMax reports that M1 outperforms some closed-source Chinese competitors in productivity tasks.

It demonstrates benchmark results that exceed those of DeepSeek’s R1-0528 model. The company also claims that M1 uses around 30% of the resources required by DeepSeek under certain conditions; however, these claims have not been independently verified.

The model was trained using large-scale reinforcement learning on 512 Nvidia H800 GPUs, with estimated rental costs of $534,700.

MiniMax has made the model open-source as part of its recent announcement.

MiniMax is among several notable Chinese AI startups, often referred to as the “Little Dragons,” supported by major firms like Tencent and Alibaba. While many competitors have shifted focus away from fundamental research due to competition from DeepSeek, MiniMax continues to develop new AI products. These include a video generation tool and an AI companion app.

.source-ref{font-size:0.85em;color:#666;display:block;margin-top:1em;}a.ask-tia-citation-link:hover{color:#11628d !important;background:#e9f6f5 !important;border-color:#11628d !important;text-decoration:none !important;}@media only screen and (min-width:768px){a.ask-tia-citation-link{font-size:11px !important;}}

🔗 Source: Bloomberg

🧠 Food for thought

1️⃣ China’s AI cost efficiency race reshapes global development paradigm

MiniMax’s claim that M1 requires only 30% of the resources used by DeepSeek represents a continuation of China’s focus on computational efficiency in AI development.

This follows DeepSeek’s own disruption of industry assumptions earlier this year, when their R1 model outperformed competitors at significantly lower costs, causing ripples through global markets including a drop in Nvidia’s stock price1.

The training of M1 on 512 Nvidia H800 GPUs at a cost of $534,700 demonstrates how Chinese companies are systematically optimizing AI training economics while maintaining competitive performance.

This efficiency-focused approach contrasts with the high-burn rate model adopted by many Western AI labs, where companies like Anthropic and OpenAI have raised billions with less emphasis on capital efficiency1.

The rapid advances in model efficiency by multiple Chinese firms indicate this isn’t an isolated achievement but rather a strategic priority within China’s AI ecosystem, potentially challenging the assumption that leading AI development requires massive capital outlays.

2️⃣ China’s “Little Dragons” illustrate the state-private partnership driving AI innovation

MiniMax’s backing from tech giants Tencent and Alibaba exemplifies China’s hybrid approach to AI advancement, where private capital combines with government strategic guidance.

The Chinese government has participated in over 60% of generative AI investment deals, creating a uniquely supportive environment for companies like MiniMax and DeepSeek to pursue fundamental AI research2.

This pattern of development follows China’s 2017 New Generation Artificial Intelligence Development Plan, which set the ambitious goal of creating a $150 billion AI industry by 2030 through coordinated public-private investments3.

Registration data reveals 3,739 generative AI tools from approximately 2,353 companies in China as of April 2025, demonstrating the breadth of the ecosystem that has developed through this coordinated approach4.

Despite earlier setbacks where many “Little Dragons” had to scale back fundamental research after DeepSeek’s rise, MiniMax’s breakthrough suggests the competitive pressure within this ecosystem continues to drive innovation rather than consolidation around a single winner.

3️⃣ The strategic shift toward long-context reasoning in AI models

MiniMax’s emphasis on its 1-million token context window (8x larger than DeepSeek R1) highlights how processing large amounts of information simultaneously has become a key competitive differentiator in AI.

This focus on extended context windows addresses a critical limitation in earlier AI systems, enabling more complex reasoning tasks across documents, code bases, and knowledge-intensive applications5.

The progression from DeepSeek’s 64,000 token window to MiniMax’s 1 million tokens in just months reflects the accelerating pace of innovation in Chinese AI, with capabilities now matching those of leading international models like Google’s Gemini 2.5 Pro6.

The competition to extend context length demonstrates how Chinese AI companies have evolved from primarily focusing on applications to now competing on fundamental technical capabilities, narrowing the gap with Western leaders in core AI technologies.

……