Personal AI launches MODEL-4, an OS for telecom operators to sell AI per line

What MODEL-4 does

According to Personal AI, MODEL-4 combines proprietary memory, unique identity, on-network inference and multi-modal agents into a service that runs on every line. The company argues that carriers are well placed to offer it because they already have a direct relationship with nearly every individual and business, a trusted identity for every line, a billing relationship and a distributed network footprint.

The release lists five core capabilities:

• Personal memory: a persistent personal model for each subscriber that remembers, learns and evolves over time, tied to the subscriber and their line.
• Network identity: the SIM and phone number serve as the agent's identity, so no separate app or login is needed.
• On-network small language models: Tier-1 SLMs handle about 95% of traffic inside the operator's network and route only complex requests to larger models.
• Voice and text agents: subscribers reach agents through native calls and texts.
• Token plans: operators can package and bill intelligence next to existing talk, text and data plans.

Suman Kanuganti, Co-Founder and CEO of Personal AI, said: "Intelligence is the next connectivity, and carriers are the ones best placed to deliver it."

Performance claims and revenue model

Building on Intelligence Everywhere orchestration research conducted with NVIDIA and Comcast, Personal AI says MODEL-4 delivers 16 ms time-to-first-token, 405 milliseconds end-to-end voice latency and a cost of $0.02 per million tokens, with operator gross margins of >90% at scale. These figures come from the company's own release.

Token plans give carriers three routes to revenue:

• Native: a token pool included in the plan powers carrier-native agents for call handling, missed calls, messaging on the subscriber's behalf and proactive reminders.
• Aggregation: for deeper reasoning or web search, the agent can route to leading LLM providers, and the carrier bills those AI platforms for access.
• Agent Store: personal agents can transact with enterprise service agents for services such as rides, reservations and banking without opening an app, letting the carrier take a share of each transaction.

Personal AI projects that a personal agent on every line can lift ARPU across an operator's entire subscriber base. It adds that because subscriber memory stays on the network under existing CPNI protections or country-specific laws, carriers can offer the services within the privacy framework they already operate.

The partner stack

MODEL-4 is designed to plug into infrastructure operators already run. NVIDIA accelerates inference, Microsoft Azure Local keeps data on-premises, Cisco optimizes the network, Rafay orchestrates deployment across core and edge, and Telegent unifies carrier voice, messaging and cellular with business communications.

On the hardware side, Personal AI says MODEL-4 uses Multi-Instance GPU partitioning on NVIDIA RTX PRO 6000 Blackwell servers with 96 GB VRAM. Each GPU slice can host up to 40 distilled small language model instances alongside ASR and TTS lanes, which the company says ensures predictable sub-second latency and zero cross-tenant interference. The platform also integrates NeMo Retriever Nemotron VL models for embeddings and reranking, vLLM-served LLMs with upgrade paths to LLM NIM, and Parakeet-0.6B for speech-to-text.

Rafay says its orchestration reduces deployment time from years to weeks. Telegent describes itself as building the first native AI mobile network platform and says carriers and MVNOs of any size can bring MODEL-4 to their customers quickly. Rodrigo Diehl, Board Member of Deutsche Telekom and CEO of Telekom Deutschland, said the network will decide where an AI workload is best executed: "on the device, at the edge, in a sovereign data center, or in the cloud."

What it means for operators and subscribers

For carriers, the pitch is a way to turn AI into a billable network service rather than leaving it to standalone apps. For subscribers, the promise is an AI that already knows them and is reached by calling or texting, with no new app to download, as Telegent CEO Spence Angerbauer put it. The release does not give pricing for token plans, named operator customers or a commercial availability date, so how quickly carriers adopt MODEL-4 remains to be seen.