OVHcloud Launches Olala mini 7A1B AI Model
The open source model was trained exclusively on European supercomputers and is already in use by French search engine Qwant.
Updated on Nov. 17, 2026 in Artificial Intelligence

OVHcloud has launched Olala mini 7A1B, an open source AI model trained exclusively on European supercomputers, including Jupiter, MeluXina, and Jean Zay. French search engine Qwant reports 20% faster responses and running costs three times lower than the model it replaced.
Why it matters
The model was developed partly in anticipation of large tech companies pulling back from open source distribution models, positioning a European-trained alternative for organizations seeking open access. Its lightweight design, which runs on a standard CPU at 50 tokens per second, could lower barriers for smaller operators.
The model has 7 billion parameters and a current context window of 32,000 tokens, running at 50 tokens per second on a standard CPU and 160 tokens per second on an Nvidia RTX A4500 GPU. OVHcloud has not yet published independent benchmarks, and the figures reflect company specifications alongside user-reported results from Qwant.
The players
OVHcloud
A European cloud computing company that develops and hosts digital infrastructure services for businesses across Europe and beyond.
Qwant
A French search engine that positions itself on user privacy and is the current user of the Olala mini 7A1B model.
The details
Olala mini 7A1B is built on a novel architecture designed to offer fast performance with lower memory requirements, and it will be distributed under an Apache 2.0 license via Hugging Face with model weights, technical specifications, and AI Act compliance documentation.
Timeline
November 17, 2026: OVHcloud launches the Olala family and Olala mini 7A1B.
Before the end of 2026: OVHcloud expects the next version to expand the context window to 128,000 tokens.
The Tech Race
The launch fits a broader contest over open-weight AI models distributed through platforms such as Hugging Face, where developers outside the largest U.S. firms are seeking alternatives. It also reflects a push to train frontier-adjacent models on sovereign European compute such as the Jupiter, MeluXina, and Jean Zay supercomputers.
Users of AI-powered search and applications may see faster responses if providers adopt lighter models like Olala mini 7A1B, which runs on standard CPUs. Organizations weighing AI costs could find that lower running expenses, as reported by Qwant, make such tools more affordable to operate.
The takeaway
Smaller, efficient open models can deliver faster responses at a fraction of the running cost, so organizations evaluating AI tools should compare speed and operating expenses alongside raw capability.
What happens next
OVHcloud expects the next version of the model to expand the context window to 128,000 tokens before the end of 2026.
Further reading
For more on how open models are reshaping the field, explore Artificial Intelligence coverage.
Source note: This article includes information reported by ITProUK.



