AIToday
Large Language ModelsAI Regulation & PolicyTop Companies' AI MovesTop Companies AIPublished: Nov 16, 2026, 06:31 JST

Digital Agency picks 7 domestic LLMs for 源内 from 15 bids

Send on LINE
Digital Agency picks 7 domestic LLMs for 源内 from 15 bids

3 Key Points

  1. What happened

    In March 2026 the Digital Agency chose 7 domestic LLMs from 15 applicants for its Genai (源内) government AI platform, with 5 built from scratch and 2 fine-tuned from overseas bases.

  2. Why it matters

    The choice marks the point where the government moves from testing to formally evaluating domestic models, and its selection criteria overlap with the security, data-compliance and safety checks companies can use when choosing a model.

Not sure about something? Ask the AI

Questions and answers are published on this page.

Summaries like this, every morning.

Context & Analysis

The selection is part of a broader push that includes GENIAC, a national project run by the Ministry of Economy, Trade and Industry and NEDO since February 2024 that provides computing resources to domestic developers. Within the seven, the mix of five scratch-built models and two fine-tuned from overseas bases reflects two development routes: building from zero lets developers control training-data rights and quality, while adapting an existing model speeds development. The government has framed domestic AI as a way to keep sensitive administrative data from depending too heavily on overseas providers, and the selection criteria mirror that emphasis on where data is handled and under whose rules. For companies, the practical read is that the government's checklist may serve as a template, though domestic models trail the largest overseas ones on top-end general performance and multilingual tasks, so mixing options by data sensitivity remains the realistic approach. Data location and the law governing it are separate questions, a distinction procurement teams may need to verify contract by contract.

FAQ
How many models applied, and how many were selected?
The public call from December 2025 to January 2026 drew 15 applications, and 7 models were selected after document screening and evaluation tests.
What happens to the selected models after the trials?
The plan calls for large-scale trials across all ministries around May 2026, use on 源内 around August 2026, partial publication of evaluation results around January 2027, and paid government procurement of top performers from April 2027 onward.
What criteria were used to judge the models?
Judging included a 50-question test shown on the day and benchmark comparisons against major overseas models, along with security for confidential information, lawful collection of training data, and explainable safety versus overseas models.
Top Companies AIRead Original Article
Send on LINE

AI news that matters for your work, delivered every morning.

Pick your industry and the AI tools you use, and get news related to your work every day.

Free · 30 seconds with Google · unsubscribe anytimeWhat is AIToday? →

Ask AI

Ask AI anything about this article. The AI reads this article, earlier AIToday articles, and Wikipedia, and cites its sources. Q&As are published on this page for other readers too.

Questions and answers are published on this page.

Related Articles

Next articleNadella: 200,000 Microsoft staff to work with 20 million AI agents