Skip to content

NewsGPU

OVH's Olala Mini Is a European Open LLM That Runs on an 8GB GPU, but Not in Ollama Yet

OVH's Olala mini 7A1B is an Apache 2.0 language model with 7B parameters but about 1B active per token, which OVH says runs on GPUs with as little as 8GB of memory. For now its GGUF files need OVH's own llama.cpp fork, and Ollama and LM Studio can't load it yet.

By , Editorial team

2 min read

OVH's Olala Mini Is a European Open LLM That Runs on an 8GB GPU, but Not in Ollama Yet

French cloud company OVH has announced Olala, its first family of large language models, and the first model, Olala mini 7A1B, is aimed squarely at modest hardware. OVH says it runs on a graphics card with as little as 8GB of memory. Forge PC hasn't tested the model; the figures below are OVH's own.

What OVH announced

OVH's press release, dated 17 November 2026, says Olala mini uses a 7-billion-parameter architecture with one billion active parameters, spread across 256 experts per layer. That mixture-of-experts design means only a fraction of the model is used for each request, which OVH says makes it faster and less memory-hungry. It combines two attention approaches (Differential-TPA and Mamba-3) and ships in BF16 format.

OVH's headline claims:

  • runs on a GPU with as little as 8GB of memory

  • 160 tokens per second on an NVIDIA RTX A4500 (a workstation card), by OVH's measurement

  • 50 tokens per second on a standard processor without a GPU, again by OVH's measurement

  • a 32,000-token context window today, with version 1.1 expected by the end of the year extending it to 128,000 tokens

OVH says the model is available as open source on Hugging Face under the Apache 2.0 licence, released with its weights, pre-trained base model and technical report; Toolhalla reports the weights went up on 19 November. OVH says it was trained on three European public supercomputers and that search engine Qwant already uses it.

The catch: app support

Toolhalla's weekly local-AI round-up (20 November) quotes the GGUF repository's warning that the files need OVH's own fork of llama.cpp, and that upstream llama.cpp, Ollama, LM Studio and other apps don't support the Olala architecture yet. Toolhalla also reports that OVH withdrew a 4-bit file after finding it measurably worse, so there's no 4-bit download for now.

That means you can't simply run ollama pull for it today. If you're new to local AI, start with a supported model in Ollama or LM Studio; our guide to running AI models locally covers the basics, and running an LLM on 8GB of VRAM explains what fits on mainstream cards.

Why it matters

OVH says the mixture-of-experts design is what makes it faster and less memory-hungry, which is exactly what mid-range GPU owners want. Olala also gives Europe a genuinely open model under a permissive licence. Once mainstream apps add support, it could be a sensible option for 8GB cards; we'll update our how much VRAM you need advice if it proves itself.

Our verdict

Olala mini 7A1B is an interesting Apache 2.0 model that OVH says runs on 8GB GPUs, but for now it needs OVH's own llama.cpp fork and doesn't load in Ollama or LM Studio. Unless you're comfortable building tools yourself, wait for mainstream app support before trying it.

The bottom line

Olala mini 7A1B is an interesting Apache 2.0 model that OVH says runs on 8GB GPUs, but for now it needs OVH's own llama.cpp fork and doesn't load in Ollama or LM Studio. Unless you're comfortable building tools yourself, wait for mainstream app support before trying it.

Sources

  1. 01OVHcloud, OVH announces Olala mini 7A1B (press release, 17 November 2026)
  2. 02Toolhalla, This Week in Local AI: llama.cpp v1.0, Olala mini, Hy4 flash

About the author

Forge PC Staff

Editorial team

The shared byline for articles researched, written and edited by the Forge PC editorial team. How we source, check and correct our work is set out in our editorial policy.

More from Forge PC Staff

Read next

How to Turn Thinking On or Off in Ollama (DeepSeek-R1, Qwen 3, gpt-oss)

How-To

How to Turn Thinking On or Off in Ollama (DeepSeek-R1, Qwen 3, gpt-oss)

Ollama lets you control reasoning models' thinking: --think and --think=false on the command line, /set think and /set nothink in chats, --hidethinking for scripts and a think field in the API. Check /api/show first, as models such as gpt-oss use low, medium and high levels instead.

Forge PC Staff2 min

How to Use Embedding Models in Ollama for Local Semantic Search and RAG

How-To

How to Use Embedding Models in Ollama for Local Semantic Search and RAG

Embedding models turn text into vectors for semantic search and RAG. Pull one such as embeddinggemma in Ollama, generate vectors with ollama run or the /api/embed endpoint, store them, then embed questions with the same model and compare using cosine similarity, all locally.

Forge PC Staff3 min

How to Turn On Variable Refresh Rate (VRR) in Windows 11

How-To

How to Turn On Variable Refresh Rate (VRR) in Windows 11

Windows 11's Variable refresh rate switch enables VRR for DirectX 11 full-screen games that don't support it natively, and Optimizations for windowed games helps windowed and borderless DX10/11 games use VRR. Turn on FreeSync or G-SYNC first; the VRR switch only appears with suitable drivers and a VRR monitor.

Forge PC Staff2 min