OVHcloud unveiled Olala, its first family of large language models, on 17 November 2026, as the French cloud group moves beyond hosting other companies’ AI. The official launch is scheduled for 19 November at the OVHcloud AI Summit at the Carrousel du Louvre in Paris.
The timetable represents an acceleration from the plans outlined in June by OVHcloud chief executive Octave Klaba, who had indicated an early 2027 launch for the company’s own models.
The first release, Olala mini 7A1B, is a compact model intended for relatively lightweight tasks. OVHcloud describes its architecture as novel and based on hybrid attention, but has published neither a complete technical specification nor benchmark results with which to assess its capabilities.
The architecture comes from Dragon LLM, formerly Lingua Custodia, a Paris start-up acquired by OVHcloud in March and renamed OVHai LLM. With Olala, the group is positioning itself as a European provider of open-source models while continuing to host those of other companies, including Chinese developers.
The aim at this stage is not to compete with the most advanced models from Anthropic or OpenAI. Olala mini instead belongs in the category occupied by Google’s Gemma, Alibaba’s Qwen and Mistral’s Ministral. Without published benchmark results, however, its standing against those competitors cannot be determined.
The only account of practical use presented comes from Qwant, which already uses the model for its flash answers. The search engine claims responses are 20% faster and costs have been cut to a third. Those figures come with a caveat: Qwant was acquired by Synfonium, a group founded by Octave Klaba. Its endorsement is therefore not that of a customer independent of his business ecosystem.
OVHcloud’s commercial strength lies in its base of 1.7 million customers and its 46 data centres. While Mistral AI has until now carried most of France’s hopes for frontier models, OVHcloud already has an extensive customer network to which it can offer its own models.
The approach echoes Google Cloud’s sale of Gemini to businesses using its infrastructure. OVHcloud could similarly offer Olala to the thousands of websites hosted on its servers, potentially at low prices, although no pricing has been specified. In the debate over AI sovereignty, that distribution capacity is a tangible advantage; it does not, however, substitute for the missing evidence about the model’s performance.
