Tencent Launches Mavis OS-Level AI Assistant

🔑 Enhanced Key Takeaways

•Mavis leverages a multi-agent architecture, allowing different AI agents to collaborate on complex tasks such as app operations, file management, and document generation, moving beyond simple text and code generation.
•The assistant operates in two distinct modes: a cloud-based efficiency mode utilizing advanced models like Tencent Hunyuan and DeepSeek V4, and a local privacy mode powered by the Qwen edge-side model.
•Mavis is capable of organizing and analyzing documents, intelligently classifying and processing images, and troubleshooting operating system issues, performing tasks typically handled by other AI agents on the market.
•It is deeply integrated with the operating system and file system, enabling it to understand device-level context, coordinate AI models based on tasks, and operate mobile applications from the desktop via cloud-based control.
•Tencent has optimized the edge-side model at the chip level, resulting in a 20% increase in running speed for local operations, enhancing efficiency for tasks directly on the computer.

📊 Competitor Analysis▸ Show

Competitor	Key Features	Pricing (Approx.)	Benchmarks (OSWorld/Terminal-Bench)
Tencent Mavis	OS-level, cross-platform (Windows, Mac, Android), multi-agent collaboration, cloud/local modes, document/image analysis, OS troubleshooting, app operation.	Not specified for Mavis, but Tencent's Yuanbao offers Hunyuan/DeepSeek models.	Not available
Microsoft 365 Copilot / Windows Copilot	AI enhancements in Office apps, free Windows 11 integration (limited capabilities), personal productivity, email management, scheduling.	$30/user/month (enterprise, on top of licenses) for M365 Copilot; Windows Copilot is free but limited.	Windows Copilot not fully autonomous agent in public build.
OpenAI Operator	Computer-use agent, advanced features via ChatGPT Plus, tool-use capability.	$200/month (via ChatGPT Pro subscription).	38% on OSWorld benchmark.
Coasty	Real computer-use agent, runs on desktops/browsers, handles CAPTCHAs, multi-window workflows, complex navigation, terminal commands.	Free.	82% on OSWorld benchmark.
Anthropic Claude Sonnet 4.6	Conversational and general-purpose AI assistant, advanced reasoning.	More than OpenAI's Operator (specific pricing not detailed for agent use).	72.5% on OSWorld benchmark.
Google Gemini	Conversational and general-purpose AI assistant, multi-language support, AI code suggestions.	Free tier available; paid from $9/month for some features.	Gemini 3.1 Pro scores 68.5% on Terminal-Bench 2.0.

🛠️ Technical Deep Dive

Multi-Agent Architecture: Mavis employs a multi-agent collaboration system, where specialized agents handle distinct tasks such as App operations, EXE operations, computer operations, file management, and document generation.
Dual-Model Operation: It offers a cloud-based efficiency mode utilizing advanced large language models like Tencent Hunyuan and DeepSeek V4. A local privacy mode is also available, powered by the Qwen edge-side model.
Chip-Level Optimization: Edge-side models have undergone chip-level optimization, leading to a 20% increase in running speed for local processing.
Operating System and File System Integration: Mavis is deeply integrated with the underlying operating system and file system, allowing it to understand device-level context and directly query and modify computer files through prompts.
Cross-Device Control: The assistant can operate mobile applications from the desktop environment through cloud-based control mechanisms.

🔮 Future ImplicationsAI analysis grounded in cited sources

Mavis will significantly enhance user productivity across Tencent's ecosystem and beyond.

Its OS-level integration and multi-agent capabilities allow for complex task automation across various platforms and applications, from work to entertainment scenarios.

Tencent's dual-mode (cloud/local) approach will attract a broader user base prioritizing both performance and data privacy.

Offering both Hunyuan/DeepSeek V4 for efficiency and a Qwen edge-side model for local, private processing addresses diverse user needs and concerns regarding data handling.

Mavis will intensify competition in the global OS-level AI assistant market, particularly against Western tech giants.

Its cross-platform availability and advanced agentic capabilities position it as a direct challenger to existing and emerging AI assistants from major tech companies like Microsoft and OpenAI.

⏳ Timeline

1998-11

Tencent founded.

2024-09

Tencent unveiled Hunyuan Turbo (MoE architecture) and an AI Coding Assistant powered by Hunyuan.

2025-07

Tencent established a '1+3+N' AI system with its self-developed Hunyuan large model as the core, including the Yuanbao AI assistant and agent development platforms.

2025-09

Tencent announced the global rollout of its Agent Development Platform 3.0 and AI Infra's 'Agent Runtime'.

2025-12

Tencent reorganized its AI research structure and appointed a new Chief AI Scientist, signaling a renewed focus on frontier models.

2026-04

Tencent launched Hy3 Preview, an open-source mixture-of-experts model with 295 billion parameters, deployed across its ecosystem including Yuanbao.

2026-05-21

Tencent officially launched Mavis, an OS-level AI assistant, across Windows, Mac, and Android platforms.

Tencent Launches Mavis OS-Level AI Assistant

⚡ 30-Second TL;DR

Key Points

🧠 Deep Insight

🔑 Enhanced Key Takeaways

🛠️ Technical Deep Dive

🔮 Future ImplicationsAI analysis grounded in cited sources

⏳ Timeline

📎 Sources (4)

👉Related Updates

PACA: Open-source tool for ancient fossil coordinate mapping

First atmosphere detected on habitable-zone exoplanet

Samsung Secures $200B Broadcom AI Infrastructure Deal

How AMD's 2006 ATI Acquisition Built Today's AI Empire