HomeArtificial IntelligenceOVH Announces Olala Mini 7A1B, a Compact AI Model That Runs on...

OVH Announces Olala Mini 7A1B, a Compact AI Model That Runs on GPUs With 8GB Memory

OVH has announced Olala, its first family of large language models (LLMs), with Olala mini 7A1B as its first release. The compact model uses a hybrid attention architecture and is designed to support agentic AI applications while requiring relatively low computing resources.

The company will officially launch the model at the OVHcloud AI Summit on November 19, 2026, where it will also showcase its latest cloud and AI developments.

OVH Olala Mini 7A1B: Architecture and Performance

Olala mini is built on a 7-billion-parameter architecture, with one billion active parameters distributed across 256 experts per layer. It combines Differential-TPA and Mamba-3 hybrid attention mechanisms and uses the BF16 format.

According to OVH, the architecture activates only a portion of the model for each request, helping reduce memory requirements and improve processing speed.

The model can run on a GPU with as little as 8GB of memory. OVH reports performance of up to 160 tokens per second on an RTX A4500 GPU. It can also run without a dedicated GPU, reaching up to 50 tokens per second on a standard processor.

Olala mini currently supports a context window of 32,000 tokens. The model will be released as open source on Hugging Face under the Apache 2.0 licence.

Trained on European Supercomputers

OVH trained Olala mini using three European public supercomputers: Jupiter, MeluXina and Jean Zay. These systems use GH200, A100 and H100 GPUs.

The training received public computing support equivalent to €7 million in computing hours. OVH said the initiative helped bring together its artificial intelligence team and establish OVH’s Agentic Lab, OVHai LLM.

The company will release the model weights, pre-trained base model, technical report and AI Act compliance documentation alongside the model.

Qwant Already Uses Olala Mini for Flash Answers

French search engine Qwant has already adopted Olala mini for its flash answer feature. According to OVH, the implementation has made responses 20% faster while reducing running costs to one-third of those associated with the previously used model.

The deployment provides an early example of how the compact model can be used in an existing AI-powered service.

Olala Mini 1.1 and 2.0: What’s Coming Next?

OVH expects to release Olala mini version 1.1 by the end of 2026. The update will extend the context window from 32,000 to 128,000 tokens and improve mathematical reasoning capabilities.

The company is also planning Olala v2.0, which will focus on three areas: enterprise agentic AI, cloud DevOps and cybersecurity.

For enterprise applications, Olala is expected to power OVHai Workspace. Its encrypted version will require models capable of running locally on workstations.

The DevOps-focused models will be designed for developers, architects and operations teams responsible for maintaining service continuity. The cybersecurity range will cover both defensive and offensive applications.

OVH says Olala v2.0 will exceed 100 billion parameters, marking a significant expansion of its AI model development plans.

Leadership Comment

Octave Klaba, CEO of OVHcloud, said: ‘AI is strategic for OVH. As well as offering AI services to our customers, we must be able to create models for our own internal needs and for those of our customers. We cannot simply rely on the generosity of certain labs to share their LLMs as open source in the years to come. Know-how in model training has become widely accessible, and the cost of training has been divided by 20 in three years. It is now possible to create LLMs while making this activity profitable.’

Also read – Shorthills AI Appoints Satyajit Bandyopadhyay as CEO and President

Join us on LinkedIn for quick updates – FYI9 News LinkedIn

LEAVE A REPLY

Please enter your comment!
Please enter your name here

Must Read