Olala mini 7A1B: OVHcloud's first LLM - a 7B MoE that activates just 1B per token
OVHcloud announced Olala, its first LLM family, starting with Olala mini 7A1B: a 7B-parameter MoE with 256 experts per layer activating only 1B per token, hybrid Differential-TPA + Mamba-3 attention, 32K context, trained on European public supercomputers and released under Apache 2.0.
Why it matters
Europe's first sovereign-scale open LLM from a major cloud provider: trained from scratch on public supercomputers (Jupiter, MeluXina, Jean Zay) with AI Act compliance docs, runs on an 8GB GPU, and Qwant already uses it for flash answers.
What you could build with it
A European startup could build a GDPR-friendly on-premise support assistant on Olala mini's 8GB-GPU footprint, running entirely inside a client's own infrastructure for regulated industries that cannot send data to US clouds, and charge per deployment rather than per token.
Does it hold up?
Too early to judge - announced Nov 17 with weights promised under Apache 2.0, but no Hugging Face page was confirmed live at announcement time and the formal launch is the Nov 19 AI Summit; no standard benchmark scores were published.
Built with Olala mini 7A1B
- OVH announces Olala mini 7A1Bpulsexpertech · Reprint of the OVH press release: 7B architecture, 1B active per token, hybrid attention in BF16, 32K context, weights plus pre-trained base model, technical report and AI Act documentation.
- AI Model Releases tracker: OVHcloud releases open-weight 7B Olala Mini LLM under Apache 2.0af.net · Live model-release tracker entry logging the Olala mini launch among the month's open-weight releases.
- November 17, 2026 AI Evening Comprehensive Reportkimkj · Evening briefing section on Olala mini: 7B MoE, 256 experts per layer, Differential-TPA + Mamba-3 attention, Apache 2.0, with the caveat that no standard benchmark scores were provided and HF availability was not confirmed for the Nov 19 event.
Learn more
First spotted on article: source.
More AI models
VoiceStudio: open-source, fully-local ElevenLabs alternative for voice cloning, dubbing, and audiobooksdebpalash/VoiceStudio is a desktop app that runs voice cloning, voice design, video dubbing, dictation, transcription…models · JEV 0.75Edge0-145B: open 145B MoE that runs on a single consumer GPUEdge0 open-sourced 145B and 120B latent-MoE models derived from Qwen3.8-Flash-Next, plus an inference framework that…models · JEV 0.74Strata: run a 125B model on a gaming PCFree, MIT-licensed local inference engine that runs Qwen3.8-Flash-Next (125B MoE, about 6B active per token) on…models · JEV 0.73Google AI Edge Gallery: offline speech recognition and synthesis on-deviceGoogle added fully offline speech recognition and text-to-speech to AI Edge Gallery, with ML Kit APIs on Android and…models · JEV 0.73Soup: fine-tune an 8B LLM on a 4 GB laptop GPU from one YAML fileAn open-source CLI that fine-tunes LLMs from a single YAML file, with layer streaming that trains an 8B model on a 4…models · JEV 0.73Tencent Hunyuan Hy3: open 295B MoE with 21B active under Apache 2.0Tencent open-sourced Hy3, a 295B-parameter mixture-of-experts model with only 21B parameters active per token, under…models · JEV 0.73
Get the week's best AI launches, plus 3 ideas worth building
One email every Saturday. Ranked by traction, not hype. Free.