# OVHcloud previews Olala mini, its first language model

Published 2026-11-18 · AI.info Team · AI.info

> OVHcloud has previewed Olala mini 7A1B, the first model in its Olala language-model family, ahead of an official launch on November 19. The company says the model will be available on Hugging Face under an Apache 2.0 licence and is already used by Qwant.

[OVHcloud](https://corporate.ovhcloud.com/asia/newsroom/news/ovh-olala-mini/)'s first language-model family will begin with a compact model that the company says can run on a standard processor without a GPU. The company introduced Olala mini 7A1B in a November 17 press release and plans to make it available as open source.

The official launch is scheduled for November 19 at the OVHcloud AI Summit, where the company also plans to present its cloud and AI offerings for customers. Olala mini is the first member of a family called Olala.

## A compact first model

Olala mini has a 7-billion-parameter architecture, with one billion active parameters spread across 256 experts per layer. Its hybrid attention combines Differential-TPA and Mamba-3, and the model uses the BF16 format.

The company says the model can run on a GPU with as little as 8 GB of memory. It reports a speed of 160 tokens per second on an RTX A4500, or 50 tokens per second on a standard processor without a GPU. The current context window reaches 32,000 tokens.

Olala mini will be available as open source on Hugging Face under the Apache 2.0 licence. The release includes the model weights, a pre-trained base model, a technical report and AI Act compliance documentation.

## Training and an early user

The model was trained on three European public supercomputers: Jupiter, MeluXina and Jean Zay. The machines were equipped with GH200, A100 and H100 GPUs.

OVHcloud values the public support at the equivalent of €7 million in computing hours and says it helped bring together the AI team behind its Agentic Lab, OVHai LLM. The company describes Olala mini as having been trained from scratch.

The company says Qwant has already adopted the model for its flash answer feature. It reports answers 20% faster and running costs divided by three compared with the model previously used by Qwant.

## Plans for later versions

Version 1.1 is expected by the end of the year. The planned update would raise the context window to 128,000 tokens and improve mathematical reasoning performance.

For version 2.0, OVHai LLM intends to develop models specialised in three areas: agentic AI for businesses, cloud DevOps and cybersecurity. The DevOps versions are intended for developers, architects and operations teams; the cybersecurity work would cover both defensive and offensive aspects.

Olala is also intended eventually to power OVHai Workspace. Its encrypted version will require models that can run locally on a workstation. The company says Olala v2.0 will exceed 100 billion parameters.

CEO Octave Klaba said OVH needs to be able to create models for its own use and for customers. “We cannot simply rely on the generosity of certain labs to share their LLMs as open source in the years to come,” he said.

## Sources

1. Original source [OVHcloud](https://corporate.ovhcloud.com/asia/newsroom/news/ovh-olala-mini/) corporate.ovhcloud.com

## Related stories

- [China Telecom details Xing 4 training on 8,192 Ascend chips](https://ai.info/news/china-telecom-xing-4-ascend-training)
- [AWS details open-source stack for Bedrock WhatsApp agents](https://ai.info/news/aws-open-source-bedrock-whatsapp-agent-stack)
- [Google plans Kubernetes Python client upgrades for AI workloads](https://ai.info/news/google-kubernetes-python-client-ai-ml-modernization)
- [llama.cpp 1.0 sets versioning rules for local AI developers](https://ai.info/news/llama-cpp-1-0-versioning-local-ai)
- [TIER IV releases driving AI models for different sensor setups](https://ai.info/news/tier-iv-open-source-driving-ai-models)

## Explore

[More articles](https://ai.info/news)

---

Canonical: https://ai.info/news/ovhcloud-previews-olala-mini-language-model
