Rapid development of AI technologies represents a major paradigm shift, impacting citizens, and societies in many ways. Yet, AI’s current trajectory faces critical economic, environmental and accessibility hurdles. Developping resilient, sustainable and efficient AI systems is key to ensuring that AI is geared to the benefit of all people and the planet.
Compression of AI models represents an important tool in developing resilient and sustainable artificial intelligence solutions. By strategically reducing model size and complexity, compression techniques enable the deployment of significantly more energy-efficient AI systems without compromising performance.
COMPETITION OVERVIEW
The governments of France and India, UNESCO and the Sustainable AI Coalition are launching the Resilient AI Challenge, an innovative competition designed to engage both the research community, industry and innovation partners. This challenge invites participants to explore novel approaches and advanced techniques for model compression, seeking the optimal balance between computational accuracy and energy gains.
Many AI models are now open source, but organizations, businesses, entrepreneurs or researchers with limited computing resources often struggle to deploy them. Making models more efficient through compression can broaden access to AI, while reducing its environmental impact. A report published, in July 2025, by UNESCO and UCL, “Smarter, Smaller Stronger, Resource Efficient AI and the Future of Digital Transformation” shows that small changes to how Large Language Models are built and used can dramatically reduce energy consumption without compromising performance. Compression is one of the technics explored.
The challenge is geared at the research community, companies and startups that whish to make the research in compression models improve.
Teams can compete in one or more of the following categories, with one AI model per category. Each model performs different tasks and is of different size to explore a variety of compressed technics :
- Audio-to-text with Voxtral Realtime by Mistral AI
- Image-to-text with Gemma 3n by Google
- Text-to-text with by Sarvam
The winner of each category will be the team that supplies the most compressed version of its category’s baseline model, meeting a threshold of accuracy and energy gains.
TIMELINE
This challenge was officially launched at the AI Impact Summit in India on February 20, 2026.
Team registration are open from February 20 to March 20, 2026, directly via the challenge webpage (see below).
The competition will officially run from March 22 to May 30, 2026.
On the week of March 22 to 27, 3 kick-off meetings will be organized (one dedicated online kick-off meeting per category). During these meetings, the project manager or tech leads for each of the model (Mistral AI, Google and Sarvam AI) will present the specific AI model that the participating teams will work with throughout the challenge.
A online Q&A session will be organized with the AI model provider mid-challenge.
Compressed models will be assessed from May 30 to July 1, 2026.
The winning teams will be announced at ITU’s annual AI for Good Summit in Geneva, Switzerland, on July 7–10, 2026.
WHY PARTICIPATE IN THE CHALLENGE
By participating in the first international competition designed to accelerate research in the field of compressed models, you will be contributing to a global effort to shape the future of AI, making it more accessible and sustainable. Join a team of trailblazers designing real AI tools and techniques that align with global environmental and developmental goals.
Prizes:
- Media exposure via UNESCO, AI for Good Summit, Sustainable AI Coalition channels…
- Visibility to the AI community with presentation at the AI Efficiency Meetup
- Presentation and discussion of the compressed models to the supporting AI model provider
- More prizes to be announced
HOW TO TAKE PART
Who can participate ?
- Researchers from universities, or research institutions
- Companies and start-ups : registered businesses, start-ups, or non-profit entities
- Working professionals employed in start-ups, companies, public sector, public sector entities or non-profits
- Students enrolled in universities or research programs
Challenges will be organized into 3 categories, to reflect various use-cases. The three categories are :
- Audio-to-text using Voxtral Realtime by Mistral AI
- Image-to-text using Gemma-3n by Google
- Text-to-text using by Sarvam
HOW THE SOLUTIONS WILL BE EVALUATED
Submissions will be evaluated by the technical team of Challenge Organizers, which will assess both the accuracy and energy gains of the compressed models. To guarantee fairness, the energy consumption of inferences will be measured under uniform conditions, using identical hardware for all evaluations.
ORGANIZED BY
WITH THE SUPPORT OF
CONTACT US
