Sophia Windsor
[ Gearbest Technology News]On May 19, 2026, Lenovo Group officially released six new and upgraded contents of Tianxi AI 4.0 full-stack technology and Lenovo AI host (AI Center) at the “AI Super Power” Tianxi AI all-in-one multi-terminal and full-scenario new product super night held in Beijing. Tianxi AI 4.0 is born for AI native devices. From +AI to AI+, from “co-pilot” to “main driver”, it achieves the key transition from passive invocation to autonomous execution of tasks for the first time, marking that personal AI has entered a new cycle of AI+. Lenovo's super products embedded with Tianxi AI 4.0 have completed the transformation from “iron to steel”, and the name “Tianxi” has been highlighted again, becoming everyone's “exclusive super partner”.
The Lenovo AI host launched simultaneously is an AI-native personal home edge device. It carries up to 190 TOPS local computing power in a palm-sized body. Together with Tianxi AI 4.0, it builds the industry's first consumer-grade end-edge-cloud integrated native AI architecture. It truly connects all devices such as AI host, AI PC, AI mobile phone, AI tablet and AIoT, making them an organic whole and building a complete AI native ecosystem.
From +AI to AI+, from “co-pilot” to “main driver”
Zhang Hua, senior vice president of Lenovo Group and general manager of China Consumer Business Group, said at the press conference that Tianxi AI 4.0 version is completing a fundamental role transformation – evolving from a passive auxiliary tool to a “personal exclusive super partner” that can independently understand intentions, independently plan paths, and autonomously perform complex tasks. Lenovo defines this upgrade as the singularity moment of “from +AI to AI+” and has established three core values for this: accuracy, security, and low cost.
Supporting this leap are the six major upgrades of Tianxi AI 4.0 full-stack technology, which integrates the full-link capabilities of human-computer interaction (+AI) and silicon-based interaction (AI+). The core breakthrough is the core execution system of the first launch – Tianxi Claw. This system builds the capability base of AI agents around four major independent innovations. The first is the bionic memory subsystem. Relying on Lenovo's self-developed 1.7 billion parameter multi-modal memory model, the system can convert multi-source information into three types of memory: situational, semantic and procedural memory, and further learn the user's personalized characteristics, ultimately forming an exclusive “personal memory map”. In the authoritative OS World global memory standard test, Tianxi Claw ranked first with a score of 74.5%, becoming the world's first intelligent memory system that surpasses 72% of human memory. In long-term complex tasks, the average token consumption is reduced by 60%, and the task execution speed is increased by more than 50%.
The second is virtual context window technology, which directly addresses the core problem plaguing the large model industry – the context window capacity is limited and large models will experience a cliff-like “intelligence reduction” after exceeding a certain threshold. Tianxi AI 4.0 first stores all historical conversations in the designated space of the device, then performs intelligent hierarchical compression management of the context, and finally accurately recalls it on demand, achieving more than 10 times expansion of the context window information carrying capacity, which is enough to support the stable execution of complex tasks that last for several days without “out of focus”.
The third is the dual-track self-evolution mechanism, which is divided into individual evolution and group evolution. At the individual level, private skills are automatically extracted from successful interactions through the Trace2Skill engine and continuously iterated. At the group level, public skills are gathered through differential privacy encryption technology to empower all users, ensuring absolute security. The fourth is full-terminal interconnection, which deeply connects more than 300 system-level atomic capabilities to achieve task relay and seamless collaboration among mobile phones, tablets, PCs and other devices.
In addition to independent execution capabilities, the second major upgrade of Tianxi AI 4.0 is the personal knowledge base, which uses a three-tier architecture to build a user's exclusive “private library”. The first layer of original document layer realizes all-formatting and intelligent deconstruction; the second layer of knowledge graph layer achieves precise positioning within seconds through a four-channel recall mechanism; the third layer of knowledge ontology layer automatically unifies semantics and directly connects to Tianxi Claw to perform complex tasks. Taking the writing of complex reports as an example, the generation time of a report with more than 10,000 words can be completed in about 8 minutes, which is shortened by 60%. It is perfectly adapted to Tianxi Claw, making Tianxi AI 4.0 not only have stronger personal exclusivity, but also have more powerful and accurate execution of complex tasks.
Building the industry's first consumer-grade end-edge-cloud integrated native AI architecture, Tianxi AI 4.0 can hold the bottom line and make progress.
The third key upgrade of Tianxi AI 4.0 is from device-cloud hybrid deployment to integrated device-edge-cloud deployment, injecting high-efficiency, low-cost and inclusive genes into it. In the current AI usage model that is common in the industry, high token fees are often prohibitive. Tianxi AI 4.0 has chosen a different path – fully upgraded from device-cloud hybrid deployment to device-edge-cloud integrated hybrid deployment, striving to achieve the goal of keeping 80% of the token consumption in AI tasks locally on the device, and only 20% completed through cloud collaboration. This is the industry’s first complete native AI architecture built for each user in the field of consumer AI, further promoting the practical implementation of inclusive AI.
The stronger the ability, the more awe is needed. The fourth major upgrade of Tianxi AI 4.0 is the trusted and secure four-layer security protection. The first layer is the behavioral control security layer, which locks private information permissions throughout the entire process, and prohibits cross-border operations; the second layer is the operating environment security layer, creating an independent and isolated exclusive “safe house” for Tianxi Claw; the third layer is the full-link data security layer, where private information is called throughout the entire process. End-to-end encryption; the fourth layer is the hardware base security layer, which builds a solid foundation of trust from the cloud to the terminal. Lenovo said that the team invested more time and energy in Tianxi’s security constraints than functional innovation, in order to make Tianxi Claw have a trustworthy soul from the bottom up.
Even if we keep the bottom line, we must also open up the boundaries. The fifth major upgrade of Tianxi AI 4.0 is the newly launched Skills Skills Plaza, which is a key step in Tianxi’s ecological expansion. Every AI developer, enthusiast and user can upload their successful practical skills to the square. The platform provides sharing space, learning space and even entrepreneurial space, allowing Tianxi AI's capabilities to continue to expand through co-creation.
Embedded with Tianxi AI 4.0, Lenovo’s many super new products have completed the transformation from iron to steel.
If these top five upgrades represent breakthroughs in native capabilities in the direction of AI+, then at the level of human-computer interaction, Tianxi AI 4.0 also brings a series of richer experiences in the +AI dimension. The two go hand in hand, and together constitute the complete capability map of Tianxi AI 4.0. This is the sixth major upgrade, which is synchronized with the new terminal products released by Lenovo at this event.
The industry's first timbre simulation AI simultaneous interpretation can reproduce personal timbres, and supports real-time multi-language translation and real-time subtitle loading in multiple scenarios; the upgraded AI Miaoji 2.0 realizes full-modal content capture and automatically generates structured minutes; the first AI smart key integrates one-click system-level entry Call out AI and solve all kinds of problems with one click; the newly upgraded AI Magic Pen allows the stylus to support intuitive interaction, directly converting reading annotation habits into instructions for controlling AI; the world's first AI smart eye function is equipped with eye tracking technology, which can realize instant AI response wherever the eyes are. In addition, I want to upgrade the AI service to version 2.0 and evolve into a home equipment manager that covers the entire life cycle of purchase, use, maintenance, repair, and replacement. All terminals are on call 24/7… More content is urgently needed to be explored.
Everyone can have an exclusive private word factory, and the AI host is specially designed for intelligent agents.
If Tianxi AI 4.0 builds a “soul” for personal AI agents, then Lenovo's AI native device simultaneously launched – Lenovo AI Host (AI Center), creates an exclusive “body” for this soul. As the industry's first personal home AI edge device, Lenovo AI host is positioned as an AI native device designed for smart bodies. It can perform tasks 24/7 and become a private “Token factory” for everyone and every family.
The first flagship model to be unveiled, the Lenovo AI host P7, attracted everyone’s attention with its huge contrast in performance and size. This palm-sized device is equipped with local AI computing power of up to 190 TOPS, can load a local large model with up to 122 billion parameters, has a maximum configuration of 80GB memory, supports 128K context windows, and has an inference speed of more than 30 Tokens per second. With such powerful performance, the maximum power consumption is only 30W. An ordinary mobile power supply can support operation, and the operating noise is controlled below 35 decibels.
Supporting this performance is the “Tianxing AI computing architecture” jointly developed by Lenovo and top laboratories. The large model enhancement engine uses exclusive sparse enhancement technology to achieve an accuracy comparable to that of a 16-bit full-precision model under 4-bit quantization; the inference acceleration engine uniformly schedules the mixed computing power of dNPU, CPU, and GPU, increasing the inference speed by 2 to 5 times under the same parameter model; the heterogeneous dynamic adaptation computing power ensures ultra-fast response, ultra-high accuracy, and ultra-long task execution capabilities. The main chip of the AI host P7 is the high-end model P1 launched by the Chinese chip company Xixin Technology. Lenovo has jointly carried out in-depth software and hardware collaborative optimization with this chip. In terms of technology, Lenovo pioneered Z-direction space multiplexing three-dimensional stacking technology, reducing the motherboard area by 40%.
Lenovo AI host P7 is also the only device in the industry that supports “one machine, two modes”. Mode 1 is the agent task execution mode, pre-installed with Tianxi Claw, which maximizes local execution of complex tasks and completely bids farewell to token consumption anxiety; mode 2 is the private token factory mode, which provides large model API keys and can be open to access any AI applications and agents, turning it into a local free token factory in seconds. When placed at home, it is a super powerful edge AI word center that can be shared by multiple family members; when put in a pocket, it is a personal pocket AI word workstation.
Also unveiled with the P7 is the Lenovo AI host mini. While the body is smaller, it still has 45 TOPS of local computing power, an average operating power consumption of only 15W, and an operating noise of less than 30 decibels. The mini is also pre-installed with Tianxi Claw, which can be used out of the box and can be teamed up with one click. It supports multi-member sharing, data isolation, independent memory, and no interference with each other.
In terms of launch pace, Lenovo AI host P7 will start crowdfunding on July 1, and the first batch will be shipped during Double 11. The final pricing has not yet been determined, but “referring to the cloud host with the same performance plus the Token subscription fee for a large model, the AI host must be more value-for-money, and the more you use it, the more money you will save.”
The six major new upgrades of Tianxi AI 4.0 are interlocking, allowing AI to truly evolve from a passively responsive “co-pilot” to a “main driver” capable of autonomous execution. Relying on this set of full-stack capabilities, Lenovo transforms the underlying advantages of equipment manufacturers into practical value that users can perceive, opening up the key link from technological breakthroughs to AI inclusiveness. “Let China lead the world in AI applications and let Tianxi AI become everyone's exclusive super partner.” Lenovo has always taken this as its direction and is firmly moving forward on the road to inclusive AI.
Login with Google