MiniMax
A Chinese AI company and model developer known for the MiniMax-M1 and M2 large language models featuring ultra-long context windows of up to 4 million tokens, strong agentic performance, and open MIT-licensed releases.
MiniMax (海螺AI, Hǎiluó AI) is a Shanghai-based Chinese AI company founded in 2021. Originally focused on multimodal AI for consumer applications — particularly AI-generated video, avatar creation, and voice synthesis — MiniMax has expanded into frontier large language model development, releasing the MiniMax-M1 and MiniMax-M2 models that have attracted significant attention for their extreme long-context capabilities and competitive benchmark performance against leading open-source models globally.
Company Background
MiniMax was founded by Yan Junjie, a former researcher at SenseTime and Carnegie Mellon University. The company built its early reputation on consumer AI products including Talkie (AI character roleplay), Hailuo AI (video generation), and an AI-powered social companion application. These products provided MiniMax with practical experience deploying large multimodal models at consumer scale, accumulating RLHF signal from millions of user interactions.
MiniMax received investment from Tencent, Sequoia China (HongShan), and Zhipu AI's founding investors, among others. In 2025, MiniMax pursued an initial public offering (IPO) on the Hong Kong Stock Exchange, one of several Chinese AI companies to list publicly as the sector matured. The IPO attracted scrutiny from analysts regarding the competitive dynamics and unit economics of the Chinese LLM market.
MiniMax-Text-01 and MiniMax-VL-01
In January 2025, MiniMax unveiled the MiniMax-01 product line, comprising MiniMax-Text-01 (a general-purpose language model) and MiniMax-VL-01 (a vision-language model with image understanding). These models were released with an emphasis on long-context processing, featuring context windows substantially larger than mainstream contemporaries.
MiniMax-M1
In June 2025, MiniMax released MiniMax-M1, positioning it as a leading open-source model for agentic applications. MiniMax-M1 features a 1 million token input context window and 80,000 token output capacity — enabling it to process books, lengthy codebases, or extensive research corpora within a single session. This context length was an industry high-water mark for open-source models at the time of release.
MiniMax-M1 was released under the MIT License, making it freely usable for commercial applications without royalties. The model demonstrated state-of-the-art performance among open-source models on agentic benchmarks measuring multi-step task completion.
MiniMax-M2 and M2.1
MiniMax-M2 (late 2025) extended the series with a hybrid Lightning Attention mechanism. Lightning Attention combines linear attention (which scales linearly rather than quadratically with sequence length) with standard softmax attention for nearby tokens, enabling native processing of up to 4 million tokens — approximately 20 to 30 times the context window of most frontier models. MiniMax-M2 can be served efficiently on as few as four NVIDIA H100 GPUs at FP8 precision, making it practical for enterprise self-hosting.
MiniMax-M2.1 refined this architecture and was described by VentureBeat as the leader among open-source models for agentic tool calling tasks, reflecting MiniMax's focus on models that can autonomously use external tools, APIs, and workflows.
Speech and Multimodal Products
Alongside its LLM work, MiniMax operates a text-to-speech product supporting over 30 languages — Speech-02 (April 2025) — which provides high-fidelity voice synthesis with emotional control and speaker cloning capabilities. MiniMax's Hailuo AI video generation product has attracted international users for short-form video creation, and the company's Talkie platform for AI character interaction has millions of users across Southeast Asia and beyond.
See Also
References
References
- The Register. (2025, June). MiniMax M1 model claims Chinese LLM crown from DeepSeek. theregister.com.
- VentureBeat. (2025). MiniMax-M2 is the new king of open source LLMs for agentic tool calling. venturebeat.com.
- Wikipedia. (2025). MiniMax (company). en.wikipedia.org.
- Intuition Labs. (2025). An overview of Chinese open-source LLMs (September 2025). intuitionlabs.ai.
- Recode China AI. (2025). Zhipu AI and MiniMax just went public. recodechinaai.com.