AnswersPublished17 sources
Lenovo previews a 300g AI host built for local models
Lenovo says the roughly 300g P7 can run models of up to 125B parameters locally, with a 256K context window. The two chip design pairs a Houmo M50 inference processor with a CIX P1 system chip.
Published byEdited with GPT-6 LunaImages generated with GPT Image 2
Research answer
Key takeaways
- Lenovo says the roughly 300g P7 can run models of up to 125B parameters locally, with a 256K context window.
- The two chip design pairs a Houmo M50 inference processor with a CIX P1 system chip.
AI Prompt
openai.comCreate a landscape editorial hero image for this Studio Global article: What did Lenovo reveal about its palm-sized, approximately 300-gram AI Host P7 preview at CSDN’s 2026 1024 Programmers’ Day and AI Builders. Article summary: Lenovo presented the AI Host P7 as a palm-sized, roughly 300-gram local AI productivity hub and opened an early-access program at CSDN’s October 24, 2026 conference, rather than announcing a fully priced retail launch. [. Topic tags: general, general web, user generated, documentation. Style: premium digital editorial illustration, source-backed research mood, clean composition, high detail, modern web publication hero. Use reference image context only for broad subject, composition, and topical grounding; do not copy the exact image. Avoid: logos, brand marks, copyrighted characters, real person likenesses, fake screenshots, UI text, readable text, watermarks,
Lenovo has shown a preview of its AI Host P7, a palm-sized device weighing about 300 grams that the company says can run large AI models locally. It was presented on October 24 at CSDN’s 2026 1024 Programmers’ Day and AI Builders Conference, alongside an early-access programme for users to try it on coding and office work. This is a preview and user-recruitment push, not a confirmed retail launch: Lenovo has yet to announce a price or general release date. 7
8
Two chips, with different jobs
The P7 pairs a Houmo M50 chip, designed for AI inference, with a CIX P1 system-on-a-chip that handles system operation and task management, according to reports. The M50’s stated performance is 160 TOPS at 10 watts. 3
15
Earlier reporting put the system’s total AI performance at 190 TOPS and its maximum power consumption at 30 watts. Those are whole-device figures, not the same measurement as the M50’s chip-level rating. 3
6
125B parameters and a 256K context window—according to Lenovo
Lenovo’s latest preview claims support for local models of up to 125 billion parameters and a context window of up to 256K, for tasks including code generation, reasoning and tool use. These are advertised maximums, not evidence that every model or task will run at those limits.
Earlier reports cited a different set of figures: 122B parameters and a 128K context window, with inference speeds also reported differently across coverage. One account gave a figure of about 50 tokens per second; another described a range of 20 to 65 tokens per second. The available reports do not provide reproducible test conditions, so it is not possible to establish how these figures compare—or whether the maximum model size, context length and speed can all be achieved together. 10
A lineup spanning text, images, documents and speech
A post-event report named Qwen3.8-Flash-Next, Qwen3.8-27B and Qwen3.6-35B-A3B among the general-purpose models supported by the preview. It also described models for image design, document recognition, speech recognition and speech synthesis. The report did not give complete details or testing conditions for each category.
Another account described mixed INT4/INT8 quantisation and said the average score across four benchmarks was 98.3% of a BF16 baseline. That is a benchmark-score comparison—not a claim that every task retains 98.3% of its results, and not a speed figure. The coverage does not provide enough information to reproduce the tests or establish that the result has been independently verified. 10
Lenovo’s pitch: a personal “Token Hub”
Lenovo is positioning the P7 as a personal Token Hub: a device that can run some models locally and connect them to coding tools or agent apps. Product information says users can configure an API or URL through a Token Manager, then view or switch installed models. The idea is to shift some repeated model use away from paid cloud services, but the published material does not quantify potential savings in an independent comparison.
Lenovo also describes built-in Tianxi AI and an automatic mode that can choose between local and cloud models based on the task. The company says the device can keep tasks running around the clock, with progress accessible from a phone, tablet or PC. These are product descriptions; the available reporting does not provide independent tests of long-running operation or the model-routing system.
Product lead Sun Ke said genuine productivity should not require people to calculate token costs before every intensive AI task—or hesitate before entrusting an important job to AI. That captures the P7’s central pitch: combine local models, task execution and connections to software tools in one personal device. 1
No price or retail date yet
At the time of the preview coverage, Lenovo had not announced the P7’s price or general release date. The early-access programme is for initial users to try the device and contribute feedback; it should not be mistaken for confirmed retail availability. 7
8
Studio Global AI
Continue your research
This page includes a source-backed answer you can continue inside Studio Global.
People also ask
What is the short answer to "Lenovo previews a 300g AI host built for local models"?
Lenovo says the roughly 300g P7 can run models of up to 125B parameters locally, with a 256K context window.
What are the key points to validate first?
Lenovo says the roughly 300g P7 can run models of up to 125B parameters locally, with a 256K context window. The two chip design pairs a Houmo M50 inference processor with a CIX P1 system chip.
Sources
aiaulis.comAI news in English - AI Aulis
intelligentliving.coLenovo AI Host P7 Runs 125B Models Locally in a 300-Gram Box
citforum.ruLenovo откроет предзаказы на компактную ИИ-систему ...
x.comLenovo Previews AI Host P7 With Houmo M50 and CIX P1 Chips, Opens Early-Access Program
post.smzdm.com联想AI主机P7定档10月24日开启预约：此芯+后摩国产芯方案曝光，本地可跑122B大模型_台式机_什么值得买