PFN Showcases PLaMo Uses at CEATEC

Preferred Networks (PFN) showcased applications built with its in-house large language model PLaMo at CEATEC 2026, held at Makuhari Messe from October 13 to 16. PFN pursues vertical integration, spanning AI semiconductors such as MN-Core, foundation models and applications. It says PLaMo is a core model developed from scratch rather than built on overseas models.

Japan’s Digital Agency has used PLaMo Translation, based on PLaMo, in its government generative AI platform Gennai since December 2025. In March 2026, PLaMo 2.0 Prime was selected as a trial model. Polimill’s QommonsAI, an administrative platform adopted by about 800 municipalities, also offers PLaMo for reading lengthy public comments and drafting replies. Dai-Ichi Hoki uses PLaMo in the beta version of ELDA, which answers questions based on legal information, and runs it on municipal premises.

Government agencies, research institutions and manufacturers have sought systems that use internal data without sending it to external cloud services. F-Secure Technologies ships GPU servers called PRIMERGY equipped with PLaMo and paired with applications for internal data.

The education service Ronjirou reviews students’ essays in about one minute and provides specific feedback. PFN’s education team developed it with teachers who coach essay writing, and launched it in March 2026 as the AI Essay Review and Exam Preparation Course. The service responds to growing demand for essay feedback as high school students have more opportunities to write essays for comprehensive selection admissions and other pathways. PLaMo assesses logical connections and the use of personal experiences according to teachers’ criteria, then checks whether students addressed earlier feedback in their revisions. The evaluation criteria and prompts for students who cannot write are designed in the application, outside the model. Since April, Shiga University has provided PLaMo Chat and PLaMo Translation to all students and faculty.

In the hiring service Thalesca AI Interview, an AI avatar talks with applicants, asks about their motivation and generates follow-up questions. After the interview, it produces an overall score, ratings by criterion, strengths and weaknesses, reasons for the ratings, and notes for the next interviewer. Okinawa Cellular Telephone and others use it for graduate recruitment. Some interviews that require large-scale processing during busy periods use overseas models such as GPT or Gemini, while PLaMo handles assessments of eight standard foundational abilities. PFN says its in-house models can be matched to tasks based on their performance, and that PLaMo is suited to assessment analysis where quality matters more than immediacy.

PFN also displayed PLaMo-VL, a vision-language model that handles images and text together. In one example, it answered “blue” when asked what color the box containing an orange object was. PFN envisions using it to recognize work procedures in factory video and report possible improvements. Its subsidiary Preferred Robotics demonstrated Kachaka Evo, an autonomous transport robot for businesses. Sales began on October 1; the robot carries up to 100 kg, moves to a designated shelf, lifts and transports a load, and returns to its starting point autonomously. PFN says it chose a method that moves under carts or shelves and lifts them because towing can damage them in factories where wheels are locked to secure loads.

Preferred Networks (PFN) showcased applications built with its in-house large language model PLaMo at CEATEC 2026, held at Makuhari Messe from October 13 to 16. PFN pursues vertical integration, spanning AI semiconductors such as MN-Core, foundation models and applications. It says PLaMo is a core model developed from scratch rather than built on overseas models.

Japan’s Digital Agency has used PLaMo Translation, based on PLaMo, in its government generative AI platform Gennai since December 2025. In March 2026, PLaMo 2.0 Prime was selected as a trial model. Polimill’s QommonsAI, an administrative platform adopted by about 800 municipalities, also offers PLaMo for reading lengthy public comments and drafting replies. Dai-Ichi Hoki uses PLaMo in the beta version of ELDA, which answers questions based on legal information, and runs it on municipal premises.

Government agencies, research institutions and manufacturers have sought systems that use internal data without sending it to external cloud services. F-Secure Technologies ships GPU servers called PRIMERGY equipped with PLaMo and paired with applications for internal data.

The education service Ronjirou reviews students’ essays in about one minute and provides specific feedback. PFN’s education team developed it with teachers who coach essay writing, and launched it in March 2026 as the AI Essay Review and Exam Preparation Course. The service responds to growing demand for essay feedback as high school students have more opportunities to write essays for comprehensive selection admissions and other pathways. PLaMo assesses logical connections and the use of personal experiences according to teachers’ criteria, then checks whether students addressed earlier feedback in their revisions. The evaluation criteria and prompts for students who cannot write are designed in the application, outside the model. Since April, Shiga University has provided PLaMo Chat and PLaMo Translation to all students and faculty.

In the hiring service Thalesca AI Interview, an AI avatar talks with applicants, asks about their motivation and generates follow-up questions. After the interview, it produces an overall score, ratings by criterion, strengths and weaknesses, reasons for the ratings, and notes for the next interviewer. Okinawa Cellular Telephone and others use it for graduate recruitment. Some interviews that require large-scale processing during busy periods use overseas models such as GPT or Gemini, while PLaMo handles assessments of eight standard foundational abilities. PFN says its in-house models can be matched to tasks based on their performance, and that PLaMo is suited to assessment analysis where quality matters more than immediacy.

PFN also displayed PLaMo-VL, a vision-language model that handles images and text together. In one example, it answered “blue” when asked what color the box containing an orange object was. PFN envisions using it to recognize work procedures in factory video and report possible improvements. Its subsidiary Preferred Robotics demonstrated Kachaka Evo, an autonomous transport robot for businesses. Sales began on October 1; the robot carries up to 100 kg, moves to a designated shelf, lifts and transports a load, and returns to its starting point autonomously. PFN says it chose a method that moves under carts or shelves and lifts them because towing can damage them in factories where wheels are locked to secure loads.