# Is Olala mini 7A1B SOTA?

Pending benchmarks. Announced Nov 19, 2026; no independent results yet.

As of Nov 22, 2026, Olala mini 7A1B is pending benchmarks on isitsota: announced Nov 19, 2026, with no independent results on the boards we track yet.

The first model in OVHcloud's Olala family: a text-only reasoning LLM for chat, instruction following and tool use, designed in France by OVHai LLM (OVHcloud group). Official model card: 6.8B total parameters and about 1B active per token, a hybrid Mamba-3 / attention mixture-of-experts architecture, and a 32K-token context window. Open weights are ungated on Hugging Face under Apache 2.0, with a stated release date of 19 November 2026. Official OVHcloud AI Endpoints serve it on a public OpenAI-compatible API as Olala-v1.0-mini-7A1B.

Canonical: https://isitsota.com/m/olala-mini-7a1b
Last updated: 2026-11-19
Tracking since: 2026-11-19

## Announced facts

| Fact | Value |
| --- | --- |
| Lab | OVHcloud |
| Announced | Nov 19, 2026 |
| API | Live (`Olala-v1.0-mini-7A1B`) |
| Open weights | Ungated open weights on Hugging Face at OVHaiLLM/Olala-v1.0-mini-7A1B. Official GGUF and MLX files are published by OVHai LLM at OVHaiLLM/Olala-v1.0-mini-7A1B-GGUF and OVHaiLLM/Olala-v1.0-mini-7A1B-MLX (OVHai LLM model card) |
| Parameters | 6.8B total / ~1B active |
| Type | Official model card: a small reasoning model for chat, instruction following, multilingual chat and tool use; the final post-trained (SFT + reinforcement learning) Olala-v1.0-mini checkpoint. It has four reasoning-effort levels and was trained with reinforcement learning to call tools and to search over many turns. Official newsroom: the first member of OVHcloud's Olala family of LLMs (OVHai LLM model card and OVH newsroom) |
| Architecture | Official model card: 36 layers, 29 Mamba-3 and 7 attention layers (one in five), each followed by a mixture-of-experts block with 256 routed experts (6 active) and 1 shared expert; only 7 of the 36 layers keep a KV cache. Official newsroom also describes hybrid attention combining Differential-TPA and Mamba-3 (OVHai LLM model card and OVH newsroom) |
| Context | Official model card: 32K tokens (input + output). A 128K-context version is described as coming soon (OVHai LLM model card) |
| Languages | Official model card: pre-trained on English and 36 other languages (38 in annealing), mostly European, plus Arabic, Chinese, Japanese and Korean (OVHai LLM model card) |
| Training | Official model card: 12.8T tokens of public datasets and the lab's own synthetic data, trained on the European supercomputers JUPITER, Jean Zay and MeluXina (OVHai LLM model card and OVH newsroom) |
| Hosted access | Official OVHcloud AI Endpoints: OpenAI-compatible API at https://oai.endpoints.kepler.ai.cloud.ovh.net/v1, model id Olala-v1.0-mini-7A1B. Official catalog: the playground can be tried with no sign-up; anonymous use is limited to 2 requests per minute, and an AI Endpoints API key raises that to 400 requests per minute. The same Endpoints service is also offered through Hugging Face Inference Providers (ovhcloud). Official catalog lists the AI Endpoints price as Free (OVHcloud AI Endpoints catalog and OVHai LLM model card) |
| License | Apache License, Version 2.0 (SPDX Apache-2.0). Licensor: OVH LLM SAS, trading as OVHai LLM (OVHai LLM model card) |

Tags marked unsure are not stated on the lab's own pages.

## What would let us call it

Verdicts use only independent boards we already track. Two consecutive snapshots are required before any crown.

- [x] Publicly announced with a source
- [x] Usable by the public (API or weights)
- [ ] Listed on an independent board we track
- [ ] Enough results for a rank in a category
- [ ] Holds across two snapshots (crowns only)

## What the lab claims

Self-reported. Shown for context; not used in any verdict.

- Lab claim: self-reported lab-run scores, with reasoning effort high, of 57.7 on IFBench, 71.2 on Multi-IF, 79.3 on LIFEBench, 50.5 on GPQA Diamond, 66.6 on TruthfulQA, 83.8 on DROP, 88.3 on HumanEval+, 75.1 on MBPP+, 60.7 on AIME 2024, 61.3 on AIME 2025, 52.7 on AIME 2026, 68.7 on MGSM, and 59.6 on BFCL V3, and an unweighted mean of 68.0 on the twelve non-BFCL benchmarks, 15.2 points above its SFT checkpoint (official model card).
- Lab claim: self-reported 160 tokens per second on an RTX A4500 and 50 tokens per second on a standard CPU, and that it runs on a GPU with as little as 8 GB of memory (official OVH newsroom and OVHai LLM announcement).

## Sources

- LAB: [Olala-v1.0-mini-7A1B (OVHai LLM model card)](https://huggingface.co/OVHaiLLM/Olala-v1.0-mini-7A1B) (2026-11-19)
- LAB: [Olala v1.0 mini technical report](https://huggingface.co/OVHaiLLM/Olala-v1.0-mini-7A1B/blob/main/Olala-v1.0-mini.pdf)
- LAB: [Olala-v1.0-mini-7A1B (OVHcloud AI Endpoints catalog)](https://www.ovhcloud.com/en/public-cloud/ai-endpoints/catalog/olala-v1-0-mini-7a1b/)
- LAB: [OVH announces Olala mini 7A1B (OVH newsroom)](https://corporate.ovhcloud.com/en/newsroom/news/ovh-olala-mini/) (2026-11-17)
- LAB: [Olala mini 7A1B: OVH launches its first LLM (OVHai LLM)](https://llm.ovhai.com/ovh-unveils-olala-mini/) (2026-11-17)
- REPO: [Olala-v1.0-mini-7A1B-GGUF](https://huggingface.co/OVHaiLLM/Olala-v1.0-mini-7A1B-GGUF)
- REPO: [Olala-v1.0-mini-7A1B-MLX](https://huggingface.co/OVHaiLLM/Olala-v1.0-mini-7A1B-MLX)

Tracking continues until an independent board lists it.

## Related

- Lab: [OVHcloud](https://isitsota.com/labs/ovhcloud.md)
- All pending releases: https://isitsota.com/releases.md
- JSON: https://api.isitsota.com/v1/models/olala-mini-7a1b
