
OVHcloud showcased Olala mini 7A1B at its November 19 AI Summit in Paris. The company had announced the model on November 17, and its endpoint catalog dates the hosted listing to November 18. The listed API model ID is Olala-v1.0-mini-7A1B.
Published specifications and access
OVHcloud describes a mixture-of-experts model that combines Mamba-3 and attention. Its API catalog gives 6.8 billion total parameters and 1.5 billion active parameters, while its summit post rounds those figures to 7 billion and about 1 billion. The catalog lists a 32K context window, tool calling, streaming and structured output. These are the provider’s specifications, and AiBrainWorX has not called the endpoint.
The summit recap says training ran on the Jupiter, MeluXina and Jean Zay supercomputers. OVHcloud also says it plans Apache 2.0 distribution on Hugging Face. The endpoint listing is public, but we could not verify a downloadable checkpoint on the official Hub profile at review time.
What the launch offers developers
The hosted listing gives builders a concrete route to evaluate the model with their own text and tools. Teams that need offline use should look for the actual artifact, license and runtime documentation. The company’s claims about running on small hardware should be checked with the intended context length and workload once the local artifacts are independently verifiable.
Primary sources reviewed. No independent hands-on tests.


