Olares One: local LLM hardware
Last updated
Olares One is a local AI box listed at 4,999 USD, with an RTX 5090 Mobile (24 GB VRAM), 96 GB of RAM and Olares OS, per Olares. We feature it as local LLM hardware for models up to about 32B shared by several people.
Olares One → Local LLM hardware → VRAM · quantization · tokens per second → The right machine for your model size
What does Olares publish about Olares One?
Olares publishes a full spec sheet for Olares One. The core parts are an NVIDIA RTX 5090 Mobile GPU with 24 GB of GDDR7 VRAM, an Intel Core Ultra 9 275HX CPU and 96 GB of DDR5-5600 RAM. They sit in a 2.15 kg box powered by a 330 W adapter, and the list price is 4,999 USD.
| Part | Olares' published spec |
|---|---|
| GPU | NVIDIA GeForce RTX 5090 Mobile, 24 GB GDDR7 VRAM, max TGP 175 W |
| CPU | Intel Core Ultra 9 275HX, 24 cores, up to 5.4 GHz |
| System RAM | 96 GB DDR5-5600 (2 x 48 GB), upgradable to 128 GB |
| Storage | 2 TB NVMe SSD, second M.2 slot free |
| Ports | Thunderbolt 5 (80 Gbps), 2.5 GbE, HDMI 2.1, USB-A 3.2 Gen 2 |
| Power | 330 W adapter, 30 W standby |
| Size and weight | 320 x 197 x 55 mm, 2.15 kg (3.1 kg with power supply) |
| Software | Olares OS, open source, built on Kubernetes |
Olares also states 1824 AI TOPS for the GPU. That vendor figure says little about LLM speed on its own.
Why is Olares One featured for local LLM hardware?
We feature Olares One because it covers the model sizes most people run locally, from 8B to 32B plus mixture-of-experts models such as Qwen3-30B-A3B, and it keeps its speed with several users. In Olares' own test with vLLM, Qwen3-30B-A3B held about 81 tokens per second per request with 8 concurrent requests.
With Ollama, the same test gave 107 tokens per second for one request and 32 per request at 8. For gpt-oss-120b, which is too big for 24 GB, Olares reports about 36 tokens per second with the model spilling into system memory. Because it's a CUDA machine, vLLM and other NVIDIA-first tools need no porting.
How do you start with Olares One?
Getting started takes one online activation and then a model install. Activation needs an internet connection, an Olares ID and the LarePass app, according to Olares' FAQ. After that the box can run entirely offline on the local network. Ollama and Open WebUI install from the Olares Market in one click each, and the Market itself needs internet.
Pull a model that fits in 24 GB at 4-bit, such as a 14B or 32B model or Qwen3-30B-A3B. Then create an account for each person, who connects with the LarePass app.
What should you confirm with Olares directly?
Ask Olares for the settings behind its published speed figures: context length, quantization and engine version. Ask which of the models you plan to run it has tested on the box. Confirm the current price too, because the shop announced a 300 USD discount per unit starting 17 Nov 2026.
Olares' own pages disagree on some details, such as which SSD slot the stock drive uses. Ask before you open the box to upgrade it.
Questions & answers
Is Olares One a GPU or a complete computer?
A complete computer. It ships with the RTX 5090 Mobile GPU, a CPU, 96 GB of RAM, 2 TB of storage and Olares OS, so it needs no host PC.
Can I add more GPU memory later?
Not inside the box. Olares says the Thunderbolt 5 port supports an external GPU enclosure, and two or more Olares devices can form a cluster. A cluster allows larger models but doesn't make a single request faster.
Sources for this entry
- Olares One product page (Company-published)
- Olares One specifications (Olares docs) (Company-published)
- Olares One FAQ (Olares docs) (Company-published)
- Olares blog: local AI hardware benchmarking (Oct 2025) (Company-published)
- Olares on GitHub (read via GitHub API 13 Nov 2026) (Company-published)
Continue this reading path
How the related websites connect
Cite this entry
Best Local LLM Hardware. “Olares One: local LLM hardware.” Content reviewed 13 November 2026.
https://bestlocalllmhardware.com/olares-one/