OVHcloud releases Olala, a small open LLM for local use
In brief
According to OVHcloud, the group presented Olala mini 7A1B on 17 November 2026, the first model in its LLM family, with an official launch on 19 November. It is a compact mixture-of-experts model (7 billion parameters, one billion active), to be released under Apache 2.0. OVHcloud says it runs on an 8 GB GPU and cites a first use at Qwant, with running costs cut by three.
Source : Next - « OVHcloud donne un avant-goût de son premier modèle, Olala »
What it changes for an SME
For an SME with 10 to 250 people, you do not need to switch tools at once. Olala mini is a first compact model, with no published quality scores and a 32,000-token context window, which is limited for long documents. If you already host open models, a European Apache 2.0 option may cut inference cost and keep data on site, noting that an 8 GB GPU would only be enough with quantisation, according to the article. Versions for agents, code or cybersecurity are undated; a 1.1 release is announced by the end of 2026 with a 128,000-token context.
What you can do
- Wait for the 19 November launch and the Apache 2.0 weights before any serious test.
- If you already run an open model locally, compare Olala mini on your own documents, not only on speed.
- Do not move a critical internal assistant until quality scores exist and the 32,000-token limit is no longer a constraint.
Let's talk about the time you could save.
In 20 minutes, we find where AI could save you time or money. Free, no commitment.