PATEO said its Thor-based AI box runs a 35-billion-parameter model without the cloud

PATEO CONNECT Technology said it had deployed NVIDIA’s Nemotron 3.5 automatic speech recognition and ByteDance’s Doubao, which it described as a 35-billion-parameter mixture-of-experts language model, on its AI Box, an add-on computing unit built on NVIDIA DRIVE AGX Thor. The company’s website lists the release under November 2, and it was also distributed through PR Newswire.

According to PATEO, the model suite also includes text-to-speech synthesis and a 2-billion-parameter language model for semantic classification and request rejection, and it can run inference locally without a cloud connection, supporting multilingual voice interaction and real-time streaming speech recognition. The company said the unit adds dedicated AI computing capacity without replacing a vehicle’s original main control chip, and described it as an upgrade path for existing vehicles whose processors cannot support larger models through software updates alone.

PATEO said it had secured a project nomination for a new-generation vehicle from a leading new energy vehicle automaker, which it did not name, with mass production for that program expected in the second quarter of 2027. It said it had announced joint development of the solution with ByteDance and an automaker earlier, at Auto China 2026, without giving a date for that announcement. It also reported a project nomination from a domestic robotics computing provider.

NVIDIA had published nothing about the deployment in its own newsroom when checked, so the description of how NVIDIA’s software is used is PATEO’s. The release gave no pricing, unit volumes or figures for the box’s computing capacity.