San Francisco, October 20th / PRNewswire / — On the eve of the NVIDIA GTC Berlin event, Personal AI released MODEL-4. The company stated that this is an operating system designed to enable telecom operators to deliver intelligence as a network service to every user line, just as they provide calls, text messages, and data. MODEL-4 integrates the computing power, network, and orchestration infrastructure required to deliver personal intelligence on a large scale for operators.

The company stated that AI is transitioning from centralized applications to a decentralized network, which represents a deliberate and decentralized shift. Operators possess unique advantages that enable them to deliver this new generation of intelligence; they have established direct relationships with nearly every individual and business around the world, possessing credible and secure identities corresponding to each line of service, as well as a distributed network footprint. MODEL-4 is not just a AI model; it also provides a missing layer of personal intelligence. It combines proprietary memory, unique identities, in-network reasoning, and multimodal agents to offer a new service delivered on each line of communication.

Personal AI, Co-Founder and CEO, stated: "Carriers connected the world through telephony, followed by SMS, and then came the internet. Intelligence is the next generation of connectivity, and carriers are best suited to provide this connection. With MODEL-4 and this ecosystem, any carrier can bring personal intelligence to any connected device."

Board member and Telekom Deutschland Chief Executive Officer Rodrigo Diehl stated: "Our network will understand context and securely connect people with machines. It will determine where the AI workloads are best executed: on the device side, at the edge, in sovereign data centers, or in the cloud. AI It won't just run on our network; it will be discovered, orchestrated, and executed on the network – in the right place, with the right performance, optimal latency, security, and the appropriate level of sovereignty."

MODEL-4: An operating system tailored for intelligent individuals on the internet

Personal AI indicates that MODEL-4 aggregates the core capabilities required for operators to provide personal intelligence through network services:

• Small language models within the network: The first-level SLM handles approximately 95% of the traffic within the operator's network, with only complex requests being routed to larger models in order to improve efficiency and economic viability.
• Token Plan: Operators can package smart services with existing call, SMS, and data plans for billing purposes.

The company stated that, based on the joint research on Intelligence Everywhere conducted with NVIDIA and Comcast, MODEL-4 has achieved a first token time of 16 milliseconds, an end-to-end voice latency of 405 milliseconds, and a cost of 0.02 US dollars per million token. In scaled-up scenarios, the operator's gross profit margin can exceed 90%. The relevant research has been published here.

Through the token plan, AI will be transformed into a network service.

Personal AI indicates that token plans to bring three new revenue streams from personal intelligence to operators:

• Aggregation: For requests that require deeper reasoning or web searches, the proxy can route them to leading LLM providers, and operators charge access fees to these AI platforms.
• Proxy stores: Individual agents can conduct transactions with enterprise service proxies, providing services such as taxi calling, booking, and banking, without the need for users to open the app, thereby allowing operators to earn a portion of each transaction.

Built for the operator ecosystem

MODEL-4 can be integrated with the technology stacks already in use by operators. NVIDIA provides inference acceleration, Microsoft Azure Local retains data locally, Cisco optimizes the network, Rafay is responsible for deployment and orchestration across core networks and edges, and Telegent unifies operators' voice, messaging, and cellular services with enterprise communications. As a result, personal intelligence is delivered as a native network service.

NVIDIA provides accelerated computing infrastructure, enabling MODEL-4 to migrate AI推理 from centralized data centers to the network, bringing it closer to where users live and work.

Personal AI indicates that MODEL-4 utilizes NVIDIA RTX PRO of 6000 Blackwell servers (96 GB of RAM) on the Multi - Instance GPU (MIG) partitions, and works closely with NVIDIA to isolate workloads; each GPU slice can support up to 40 distilled small language model instances, as well as ASR and TTS channels, to ensure predictable sub-second latency and zero cross-tenant interference. On this hardware foundation, the platform integrates components of NVIDIA, such as NeMo Retriever Nemotron VL models for dense embedding and reordering (llama-nemotron-embed-vl-1b-v2 and llama-nemotron-rerank-vl-1b-v2), LLM which provides services through vLLM and is upgradable to LLM NIM, as well as Parakeet-0.6B for speech-to-text processing.

NVIDIA Telco, Vice President of Global Business Development, stated: "Voice is the most native application for telecom operators. The personalized voice services provided by Personal AI represent a natural extension, opening up new opportunities for consumers, businesses, and the telecommunications industry. NVIDIA's AI-RAN platform offers a distributed computing layer that delivers new AI native voice applications in the most efficient manner."

Microsoft Azure Local extends the capabilities of Azure to the infrastructure within operators' own networks. By running MODEL-4 on Azure Local, operators can retain subscribers' memories and data locally, and keep them under the same control as information of other subscribers.

Microsoft WW Telco and Media Industry CTO Rick Lievano indicate that "telecommunications operators are increasingly hoping to run critical AI workloads in locations closer to where the data is generated, while still maintaining their expectations for security, control, and consistent management. Azure Local brings the capabilities of Azure directly into the operators' distributed infrastructure, allowing solutions such as Personal AI's MODEL-4 to run locally, while maintaining a consistent Azure experience in their respective environments."

Rafay provides infrastructure orchestration, while MODEL-4 ensures deployment and operation in operators' networks and edge environments, reducing deployment time from several years to just a few weeks.

Rafay, Co-Founder and CEO, stated: "Operators should not spend several years putting AI into production. The orchestration of Rafay with MODEL-4 has shortened this process to just a few weeks. We are proud to collaborate with Personal AI to deliver fast intelligence on operators' networks."

Cisco provides mobility and network infrastructure deployed in the networks of many of the world's largest operators, and optimizes these networks for a new category of traffic – intelligence.

Cisco, the Chief Architect, Provider Mobility, the person in charge, and Bob Everson stated: " Cisco is already at the core of global operators' networks. Together with Personal AI, we are actively exploring how to adjust the network to support distributed intelligence, enabling operators and partners to provide AI with the performance and reliability that subscribers expect."

Telegent is building the first native AI mobile network platform, which unifies the voice, messaging, and cellular services of American operators with the communication services already in use by enterprises.

Telegent CEO Spence Angerbauer stated: "Mobile phone numbers remain the most trusted form of identification for people, and calls and text messages are still the ways they communicate with important individuals and businesses. Together with Personal AI, we are making this form of identification more intelligent, so that each subscriber can call or send text messages to a AI that already knows them, without the need to download any new apps. Since Telegent makes mobile networks programmable, operators of any size and MVNO can quickly provide this feature to their customers."

Personal intelligent operating system

The definition of the next generation of AI will not be about models becoming smarter, but rather about how intelligence can become more personalized.

MODEL-4 provides operators with an operating system that is used to deliver personal intelligence as a network service: this intelligence is capable of memorizing, learning, and maintaining context between different experiences, becoming more personalized and valuable over time.

Media contacts: Amanda Borgiotti, email omitted; Jonathan Bikoff, email omitted