🔥36氪•Stalecollected in 16m
MiniMax files trademark for new AI brand Mavis
💡Track potential new model releases from a leading Chinese AI unicorn.
⚡ 30-Second TL;DR
What Changed
MiniMax filed for 'MINIMAX MAVIS' trademark
Why It Matters
The trademark filing suggests MiniMax is preparing to launch a new AI product or service line, likely expanding their LLM ecosystem.
What To Do Next
Monitor MiniMax's official channels for the upcoming release of Mavis to evaluate its capabilities against existing LLMs.
Who should care:Developers & AI Engineers
Key Points
- •MiniMax filed for 'MINIMAX MAVIS' trademark
- •Trademark covers scientific instruments and website services
- •Company specializes in AI foundation software and application development
🧠 Deep Insight
Web-grounded analysis with 17 cited sources.
🔑 Enhanced Key Takeaways
- •The 'MINIMAX MAVIS' trademark filing is associated with the rebranding of MiniMax's desktop Agent product, which now features multi-agent team collaboration capabilities.
- •MiniMax, officially Shanghai Xiyu Jizhi Technology, was founded in December 2021 by former computer vision researchers from SenseTime.
- •The company has developed a suite of multimodal AI models (text, audio, image, video, and music) and consumer applications, including the AI character apps Talkie and Xingye, and the video-generation service Hailuo AI.
- •MiniMax successfully completed its initial public offering (IPO) on the Hong Kong Stock Exchange in January 2026.
- •The company has secured substantial funding, including a $600 million round led by Alibaba Group in March 2024, valuing it at $2.5 billion, and a $300 million Series B extension in July 2025, which raised its valuation to $4 billion with investment from Shanghai state-owned capital.
📊 Competitor Analysis▸ Show
Competitor Analysis: MiniMax in the Chinese AI Landscape
MiniMax operates within a highly competitive Chinese AI market, facing both specialized AI startups and vertically integrated tech giants. The company's offerings, particularly its foundation models and consumer applications, place it in direct competition with several prominent players.
| Feature/Category | MiniMax | DeepSeek | Zhipu AI | Alibaba (Qwen) | Tencent (Hunyuan) | ByteDance (Doubao) |
|---|---|---|---|---|---|---|
| Primary Focus | Multimodal AI models, consumer apps (Talkie, Hailuo AI), enterprise APIs, AI agents | Open-source LLMs, cost-efficient models | ChatGLM series, open-source community focus | LLMs for enterprise, cloud platform integration | Video generation, reasoning AI models, social/gaming ecosystems | Consumer AI apps, spatial models, TikTok integration |
| Key Models/Products | MiniMax M2.7, Hailuo 2.3, Speech 2.8, Music 2.5+, MiniMax-01 series (Text-01, VL-01), Mavis (Agent) | DeepSeek-R1, DeepSeek-V3 | ChatGLM series | Qwen models | Hunyuan | Doubao |
| Model Architecture | Lightning Attention, Mixture of Experts (MoE) (32 experts, 456B total params, 45.9B activated per token for MiniMax-01) | Deep learning, large-scale language models | - | - | - | - |
| Context Window | MiniMax-Text-01: up to 1M tokens training, 4M inference (20-32x longer than GPT-4o/Claude-3.5-Sonnet) | - | - | - | - | - |
| Pricing (M2.5) | Input: $0.15/M tokens; Output: $1.20/M tokens. M2.5 Lightning: $0.30/$2.40/M tokens. Enterprise AI agent: ~$0.15/task. | Low-cost (compared to ChatGPT) | - | API pricing cut to $0.02/M tokens (ByteDance) | - | API pricing cut to $0.02/M tokens |
| Valuation/Funding | $4B (July 2025) | - | One of China's "Four Little Dragons" in AI | - | - | - |
| Noteworthy | Strong in multimodal capabilities, ultra-long context processing, agentic performance. IPO in Jan 2026. | Open-source strategy, cost advantages, comparable to ChatGPT performance. | Significant traction in open-source community. | Leading big tech AI champion in China, 90,000+ enterprise users. | Strong distribution via WeChat and gaming. | Leverages TikTok's user base, strong in consumer AI apps. |
Note: Pricing and benchmark data for competitors are less consistently available or directly comparable across all sources. The table focuses on available information to provide a general overview.
🛠️ Technical Deep Dive
- MiniMax-01 Series: Includes MiniMax-Text-01 (general-purpose LLM) and MiniMax-VL-01 (vision-language model).
- Core Architecture: Utilizes "lightning attention" for efficient scaling and integrates a Mixture of Experts (MoE) architecture.
- MoE Configuration: Features 32 experts with a total of 456 billion parameters, where approximately 45.9 billion parameters are activated for each token processed.
- Context Window: MiniMax-Text-01 can handle up to 1 million tokens during training and extrapolate to 4 million tokens during inference, offering a significantly longer context window (20-32 times) compared to models like GPT-4o and Claude-3.5-Sonnet.
- Vision-Language Training: MiniMax-VL-01 was built through continued training with 512 billion vision-language tokens.
- MiniMax-M1 Model: An open-source model (Apache 2.0 license) with 456 billion parameters (45.9 billion activated per token), built on the MiniMax-Text-01 foundation. It was trained using a new Reinforcement Learning (RL) algorithm called CISPO, which significantly accelerated the learning process.
- Mavis Agent Architecture: The rebranded Mavis desktop Agent product employs a code-based state machine for multi-agent collaboration, defining distinct roles such as Owner, Worker, and Verifier to manage complex, long-duration tasks and ensure task decomposition, execution, and validation.
🔮 Future ImplicationsAI analysis grounded in cited sources
MiniMax will intensify its focus on AI agent technology and multi-agent systems.
The rebranding of its desktop Agent product to Mavis and the introduction of multi-agent team collaboration indicate a strategic shift towards more sophisticated, autonomous AI solutions for complex tasks.
MiniMax is likely to expand its presence in the AI hardware and robotics sectors.
The establishment of Guangzhou Xiyu Jizhi Technology, a wholly-owned subsidiary focused on intelligent robots, AI hardware sales, and system integration, signals a move towards a software-hardware integrated strategy.
The company will streamline its product offerings and subscription models for developers and enterprises.
The unification of previously separate API and Agent subscriptions into a single plan, integrating CLI, API, and Agent functionalities under a unified quota, suggests an effort to simplify access and encourage broader adoption of its AI ecosystem.
⏳ Timeline
2021-12
MiniMax founded by former SenseTime researchers.
2022-10
Launched its first product, Glow, an AI companion app.
2023-06
Launched Talkie, the international counterpart to Glow.
2024-03
Alibaba Group led a $600 million financing round, valuing MiniMax at $2.5 billion.
2025-01
Unveiled the MiniMax-01 LLM product line, including MiniMax-Text-01 and MiniMax-VL-01.
2026-01
MiniMax held its Initial Public Offering (IPO) on the Hong Kong Stock Exchange.
2026-01
Established Guangzhou Xiyu Jizhi Technology for AI hardware and robotics.
2026-05
Rebranded its desktop Agent product to Mavis, introducing multi-agent collaboration.
📎 Sources (17)
Factual claims are grounded in the sources below. Forward-looking analysis is AI-generated interpretation.
📰
Weekly AI Recap
Read this week's curated digest of top AI events →
👉Related Updates
AI-curated news aggregator. All content rights belong to original publishers.
Original source: 36氪 ↗