Open Source

OVHcloud Unveils Olala Mini: 7B Model, 1B Active

November 17, 2026 · By Editorial Team

OVHcloud Unveils Olala Mini: 7B Model, 1B Active

OVHcloud has unveiled Olala, its first family of large language models, starting with a compact model that punches well above its size. Announced on 17 November, Olala mini 7A1B packs 7 billion parameters but activates only 1 billion of them for each token, using a mixture-of-experts design with 256 experts per layer.

The model combines two attention mechanisms, Differential-TPA and Mamba-3, and was trained from scratch on three European public supercomputers: Jupiter, MeluXina and Jean Zay. It ships with a 32,000-token context window, with version 1.1 due by the end of the year to extend that to 128,000 tokens and improve mathematical reasoning.

Efficiency is the headline. OVHcloud says Olala mini runs on a graphics card with as little as 8 GB of memory, reaching 160 tokens per second on an RTX A4500, and manages 50 tokens per second on a standard processor with no GPU at all. That makes it a practical option for running capable models locally, for example through Ollama, rather than renting cloud compute.

The release is open source under the Apache 2.0 licence, with weights, a pre-trained base model, a technical report and EU AI Act compliance documentation published through Hugging Face. French search engine Qwant has already adopted the model for its flash-answer feature. One caveat: OVHcloud has not published standard benchmark scores yet, so independent results will matter more than the vendor’s speed claims. The official launch takes place at the OVHcloud AI Summit on 19 November.

Source: OVHcloud, Unite.AI

Related stories