HaiAI123

Curated Global AI Tools Directory

Industry update
Flat

OVHcloud Open-Sources Its First AI Model Olala mini 7A1B

11/17/2026 — 11/17, 19:19·2 sources (all)·2 reports (all)

Key takeaways

  • European cloud provider OVHcloud has introduced the Olala AI family and open-sourced its first member, Olala mini 7A1B. Featuring a 7B param…

Background & analysis

On November 17, 2026, European cloud provider OVHcloud announced the launch of its first AI model family, Olala, and simultaneously open-sourced the family's first member, Olala mini 7A1B. According to IT Home's report that day, the model combines Differential-TPA and Mamba-3 in a hybrid attention mechanism, carries 7 billion total parameters with 1 billion active, uses BF16 format, and is released under the Apache 2.0 license on Hugging Face. Its initial release supports a 32K context.

On deployment and training, OVHcloud states that Olala mini 7A1B can run on a GPU with as little as 8GB of memory, reaching 160 tokens per second on an RTX A4500, and can also manage 50 tokens per second on a CPU-only setup. OVH adds that the model was trained on three European public supercomputers using NVIDIA's GH200, A100, and H100, with total costs reaching 7 million euros. The report also notes that version 1.1, expected by the end of the year, will extend the context to 128K and deliver stronger math capabilities.

The same day, the Reddit community LocalLLaMA gave additional details, describing the model as a 7B architecture with 256 experts per layer, a MoE setup, 1 billion active parameters, the same hybrid Differential-TPA and Mamba-3 attention in BF16 format, and per-request activation of only a fraction of the model, which makes it faster and less memory-hungry. Its 160 tokens per second on an RTX A4500 matches the earlier report. The two accounts align on parameters and runtime figures, and the event now stands at the stage where the model is open-sourced and the follow-up version remains to be released.

AI-generated from 2 reports · updated 11 hours ago

Latest turnOVH has announced Olala mini 7A1B, a 7-billion-parameter hybrid model featuring 256 experts per layer and Differential-TPA combined with Mamba-3 attention. It runs on an 8GB VRAM GPU and achieves 160 tokens per second on an RTX A4500.

updream
24-hour heatHeat index 140 · peak 195 · 11h ago
24 hours agonow

Reports on this story headlines open the original

Yesterday
  1. OVH has announced Olala mini 7A1B, a 7-billion-parameter hybrid model featuring 256 experts per layer and Differential-TPA combined with Mamba-3 attention. It runs on an 8GB VRAM GPU and achieves 160 tokens per second on an RTX A4500.

    Reddit · LocalLLaMAAI score 59
  2. European cloud provider OVHcloud has announced Olala, its first family of AI models, releasing the initial member Olala mini 7A1B as open source under the Apache 2.0 license. Featuring a hybrid architecture combining Differential-TPA and Mamba-3 with 7B total and 1B active parameters, the model runs on 8GB VRAM GPUs and supports a 32K initial context window.

    IT之家AI score 65

Other stories people are talking about

How is heat calculated?About the method

Heat counts how many independent sources covered a story in the last 48 hours: one source counts once no matter how many posts it published, decaying with a 24-hour half-life. What ranks first is what many people are talking about.

This page aggregates public feeds. Headlines and summaries are machine-organized and remain the property of the original authors; verify important facts at the source. If you believe a headline or summary infringes your rights, email the contact address in the footer with the page URL and basis of your claim, and we will remove or replace it after review.

Surge
Discussion rising fast
New
First report within 6 hours
Rising
Still gathering discussion

Back to AI industry updates →