Two hundred people in Asia-Pacific by 2027. Mistral AI set this objective by announcing, on November 20, 2026, the opening of its regional headquarters in Singapore. Billed as an extension of the company’s science hub, the office will develop models for the public sector, high-tech industry and robotics. This is more than double the current workforce in the region.
Arthur Mensch, co-founder and CEO of Mistral, sees this as a long-term commitment “to building sovereign AI capabilities in the region, for the region”. Governments will be able to adapt and run its open weight models in their own environments, without their data leaving a defined perimeter, the company says.
Mistral has been working with Singaporean Defense and Interior Services since March 2025
The public sector is the most advanced of the three areas. The office relies on two public partnerships. In March 2025, the Singaporean Ministry of Defense, its technology agency DSTA and DSO National Laboratories announced that they were developing generative AI models with Mistral for decision support for the armed forces. DSO, the nation’s leading defense research organization, wanted to train its own multilingual models and run them on its sites. The two teams co-developed, with linguistic data obtained from AI Singapore, an expert mixture model, an architecture that activates only part of its parameters for each element processed. The second phase concerns small reasoning models. They are intended for autonomous systems deployed in the field.
HTX, the science and technology agency of the Ministry of the Interior, which works in particular for the police, the prison administration and the immigration services, has been collaborating with Mistral since March 2025. It wanted models which understand the Singaporean context and operate on an infrastructure isolated from the Internet. Out came Phoenix. The Small and Medium versions of this series of language models, pre-trained on the basis of Mistral’s models and then refined on HTX data, run on NGINE, the agency’s AI infrastructure. Introduced in April, Phoenix-VL 1.5 Medium processes text and images with 123 billion parameters.
A memorandum of understanding signed in November 2025, then a framework agreement concluded on April 28, 2026, extended this work to robotics, video analysis and cybersecurity. In the spring, Singtel, NCS and ST Engineering also signed memorandums of understanding with Mistral.
Mistral’s first navigation model guides a robot with a single camera
In the high-tech industry, Mistral wants to accelerate engineering simulations and optimize processes. In robotics, the press release highlights Robostral Navigate, presented on July 8, 2026. This model with 8 billion parameters, designed in-house and trained entirely in simulation, drives a wheeled, legged or flying robot based on instructions in natural language and images from a single color camera, without a depth sensor. It indicates in the image the next point to reach. To access it, Mistral refers to its teams.
In June, the DSTA extended its research with Mistral to the navigation of drones guided by vision-language models. HTX, for its part, expects robots that understand orders in natural language and find their way around the city, in unknown places. For Pee Beng Kong, executive vice-chairman of the Economic Development Board, Singapore’s economic development agency, this investment will create “opportunities for Singaporeans to work on cutting-edge AI technologies.”
Singapore headquarters arrives less than two months after Munich hub
On September 28, Mistral opened a hub in Munich where research teams in physical AI and industrial AI work, as well as engineers serving client companies. Deutsche Telekom announced on November 19 that it had invested 100 million euros in the company, which had raised 3 billion euros in September in a round led by Korean Samsung Electronics.
The Singapore project dates back to January 2025. Mistral then planned to open an office there to cover the Asia-Pacific, reported Bloomberg, taken up by TechCrunch.
Our articles will be prioritized in Google’s “Featured” section.

