MiniMax Unveils M2 Model to Compete on Speed and Cost
00:00/00:00
Listen to this article
1x
Subscribe now
Listen to the full version
Listen to the full version
Despite its technological advances, MiniMax is facing legal challenges abroad. Photo: VCG
Chinese AI startup MiniMax on Oct. 27 released a new open-source inference model designed to make the development of AI agents more efficient and affordable, joining a fierce price war in the country’s large model sector.
The new model, M2, is built on a Mixture-of-Experts (MoE) architecture with 230 billion total parameters and is tailored to balance performance, cost, and speed for developers, the company said. It aims to address a common industry challenge where high-performing overseas models are too costly and slow for mass-market agent applications, while more affordable domestic alternatives often lack sufficient capabilities.
You've accessed an article available only to subscribers
VIEW OPTIONS
Unlock exclusive discounts with a Caixin group subscription — ideal for teams and organizations.
Subscribe to both Caixin Global and The Wall Street Journal — for the price of one.
- RELATED
Share this article
Open WeChat and scan the QR code
- DIGEST HUB
Digest Hub
Back
Explore the story in 30 seconds
- MiniMax launched the M2 AI model with 230B parameters, targeting affordability and speed for AI agents, priced at 2.1 yuan/million tokens (input) and 8.4 yuan/million tokens (output).
- M2 outpaces Anthropic’s Claude Sonnet 4.5 in speed and cost but slightly lags in programming; its context window is 204,800 tokens, down from M1’s 1 million.
- MiniMax, serving 157M users in 90+ regions, faces a US copyright lawsuit from major studios over its Hailuo AI platform.
AI generated, for reference only
Who’s Who
- MiniMax
- MiniMax is a Chinese AI startup founded in 2022. They recently released M2, an open-source inference model designed to make AI agent development more efficient and affordable, featuring a Mixture-of-Experts architecture. While they provide various AI products, including Hailuo AI (text-to-image/video) and MiniMax Audio, they are currently facing a copyright lawsuit from Disney, Universal, and Warner Bros. over alleged intellectual property misappropriation by Hailuo AI.
- Anthropic
- Anthropic is a company with an AI model called Claude Sonnet 4.5. MiniMax claims its M2 model is nearly twice as fast as Claude Sonnet 4.5 at only 8% of the cost. In company tests, M2's performance in tool use and deep search tasks ranked between OpenAI's GPT-5 and Anthropic's Claude Sonnet 4.5, although M2 slightly lagged behind both in programming capabilities.
- DeepSeek
- DeepSeek initiated an impressive price war in the large model sector, significantly reducing API call prices. The startup cut input costs from 0.5-4 yuan to 0.2-2 yuan per million tokens and output costs from 12 yuan to 3 yuan. This move in May 2024 led other major players like ByteDance, Alibaba, Google, and OpenAI to follow suit, intensifying competition in the AI market. DeepSeek-V3.2-Exp's pricing remains competitive with MiniMax's new M2 model.
- Baidu
- Baidu X1 is mentioned as a competitor in the landscape of large model services in China, with its pricing comparable to MiniMax's new M2 model. The article doesn't provide further details about Baidu's specific functionalities or competitive strategies in this context.
- Tencent
- Tencent is mentioned in the article in the context of pricing comparison for AI models. Specifically, its Tencent Hunyuan T1 model is noted to have a lower cost than MiniMax's new M2 model. M2's pricing is twice that of Tencent Hunyuan T1.
- Kimi
- Kimi K2's pricing is double that of MiniMax's new M2 large language model. This places Kimi as a more expensive option compared to M2 in the ongoing price war within China's large model sector.
- OpenAI
- In company tests, MiniMax's M2 model's performance in tool use and deep search tasks ranked between OpenAI's GPT-5 and Anthropic’s Claude Sonnet 4.5. However, it slightly lagged behind OpenAI's offering in programming capabilities. OpenAI was also mentioned as having followed suit in an earlier round of price cuts for large model services.
- ByteDance
- ByteDance was involved in a price war for large model services, initiated by DeepSeek in May 2024. This led to ByteDance, along with other tech giants like Alibaba, Google, and OpenAI, adjusting their pricing strategies.
- Alibaba
- Alibaba is mentioned in the article as one of the companies that followed DeepSeek's earlier round of price cuts in May 2024 for large model services. This indicates Alibaba's participation in the competitive pricing landscape within the AI sector.
- In May 2024, Google, alongside other tech giants like ByteDance, Alibaba, and OpenAI, responded to a round of price cuts initiated by DeepSeek in the large model sector's ongoing price war.
- SenseTime
- SenseTime is mentioned as the former employer of Yan Junjie, the founder of MiniMax. Yan Junjie was previously a vice president at SenseTime before establishing MiniMax, an AI startup playing a role in China's large model sector.
- Disney
- Disney, along with Universal Pictures and Warner Bros. Discovery Inc., filed a joint copyright lawsuit against MiniMax in a California district court on September 16. The lawsuit alleges that MiniMax's Hailuo AI platform infringed on their intellectual property by allowing users to generate and download images and videos of copyright-protected characters using simple text prompts.
- Universal Pictures
- Universal Pictures filed a joint copyright lawsuit against Chinese AI startup MiniMax in a California district court on September 16. The lawsuit, also involving Disney and Warner Bros. Discovery Inc., alleges that MiniMax's Hailuo AI platform misappropriated their intellectual property by allowing users to generate and download images and videos of copyright-protected characters.
- Warner Bros. Discovery Inc.
- Warner Bros. Discovery Inc. is referenced in a joint copyright lawsuit filed against MiniMax in a California district court on September 16. The lawsuit, also involving Disney and Universal Pictures, alleges that MiniMax's Hailuo AI platform misused their intellectual property by allowing users to create images and videos of copyrighted characters using text prompts.
AI generated, for reference only
What Happened When
- May 2024：
- DeepSeek initiated an earlier round of price cuts for its large model services, prompting competitors including ByteDance, Alibaba, Google, and OpenAI to follow suit.
- June 2024：
- MiniMax launched its M1 model, which was promoted for its 'long-text understanding' with a market-leading 1 million-token input capacity.
- Sept. 16, 2024：
- Disney, Universal Pictures, and Warner Bros. Discovery Inc. filed a joint copyright lawsuit against MiniMax in a California district court.
- Sept. 29, 2024：
- DeepSeek slashed its API call prices, reducing input and output costs for large model services.
- Oct. 27, 2024：
- MiniMax released its new open-source inference model, M2, designed to improve efficiency and affordability in AI agent development.
AI generated, for reference only
- PODCAST
- MOST POPULAR
Sign in with Google
Sign in with Facebook
Sign in with 财新