Buy If / Skip If
Quick decision guide — does the Olares One fit your needs? PromptQuorum's reading of the published specifications, not a hands-on test.
- Buy the Olares One if:
- You want 24 GB of GDDR7 VRAM in a small desktop (320 × 197 × 55 mm, 2.15 kg) instead of building a tower
- You want a packaged local-AI operating system (Olares OS) rather than assembling drivers and runtimes yourself
- You mainly run models up to roughly 30B at 4-bit, where everything fits in VRAM
- You value low idle noise (Olares publishes 19 dB at idle)
- Skip the Olares One if:
- Your main goal is fast 70B inference — 24 GB of VRAM forces partial offload to system RAM
- You want the lowest price per GB of VRAM — a used desktop GPU usually costs far less, at the price of size and noise
- You need independently measured benchmarks before buying — Olares' page publishes none, and we have not tested a unit
Verdict
The Olares One is a well-specified, compact box for people who want a ready-made local-AI desktop and accept a $4,999 price. Its 24 GB of GDDR7 VRAM suits mid-size models and fast prompt processing; its 96 GB of DDR5 adds capacity for 70B models, but capacity is not speed — layers held in system RAM run at DDR5 bandwidth. The strongest case for it is convenience and size, not price per GB of VRAM. We have not tested a unit, so treat this as a reading of published specifications, and treat any benchmark figure — Olares' or ours — as unverified until an independent test confirms it.
No Affiliate, No Commission
This is not an affiliate page. PromptQuorum has no affiliate agreement with Olares, receives no commission, fee or payment for this page, and adds no affiliate parameters to any link. The Olares One page was written after Olares contacted us with product information; Olares itself calls its price offer a "partner quote", but that is Olares' wording and does not describe a partnership with PromptQuorum.
- Links to olares.com and shop.olares.com are plain reference links that earn PromptQuorum nothing, whether or not you buy
- Figures sent to us by Olares are labelled as vendor claims wherever they appear
- Where a vendor claim conflicts with Olares' own product page, we show both and say which one is published
Olares One Specifications
Manufacturer-published specifications, taken from olares.com/one on 2026-10-31. Olares publishes no retail price on that page; the price comes from the official shop.
Component | Specification (Olares) |
|---|---|
| CPU | Intel Core Ultra 9 275HX, 24 cores, up to 5.4 GHz, 36 MB cache |
| GPU | NVIDIA GeForce RTX 5090 Mobile, 24 GB GDDR7, 175 W max TGP |
| Memory | 96 GB (2×48 GB) DDR5-5600, upgradable to 128 GB |
| Storage | 2 TB NVMe SSD (PCIe 4.0), second PCIe SSD slot |
| Ports | 1× Thunderbolt 5, 1× 2.5 GbE RJ45, 1× USB-A 3.2 Gen2, 1× HDMI 2.1 |
| Wireless | Wi-Fi 7, Bluetooth 5.4 |
| Size and weight | 320 × 197 × 55 mm (about 3.5 L), 2.15 kg (3.1 kg with power supply) |
| Noise (Olares lab figure) | 19 dB idle; 38.8 dB at full power (GPU 175 W + CPU 55 W), measured at 25 °C and 0.5 m |
| Power | 330 W power supply; 30 W standby; GPU 95 W in Silent Mode, 175 W in Performance Mode |
| Software | Olares OS (described by Olares as open source), free system updates |
Vendor Claims Checked
Olares sent us these figures by email on 2026-10-31. PromptQuorum has not reproduced any of them. The table sets each claim against what Olares publishes on its own product page and against PromptQuorum's estimate.
Vendor claim (email) | Olares product page | PromptQuorum status |
|---|---|---|
| Llama 3 70B (Q4_K_M): about 18–22 tokens/s sustained | No tokens-per-second figure published | Unverified. A Q4_K_M 70B model is larger than 24 GB of VRAM, so part runs from DDR5 system RAM; our estimate in the 70B guide is about 2–4 tokens/s |
| "96 GB unified memory" with the RTX 5090M | 96 GB DDR5 system RAM plus separate 24 GB GDDR7 VRAM | The product page describes separate memory pools, not unified memory |
| Under 32 dB at full continuous load | 19 dB idle, 38.8 dB at full power, under 39 dB at maximum load (25 °C, 0.5 m) | Conflicts with the published figure; we use the product page number |
| deepseek-v4.1-flash-10.17 multi-unit deployment at sub-millisecond local latency | No such model or latency figure published | Unverified: we could not identify the model name, and sub-millisecond latency is not a meaningful figure for text generation |
| $4,959 "partner quote" ($40 off $4,999) | Not listed; shop price $4,999 | The offer link Olares supplied lists $4,949 and is titled for another audience, so we do not link it (checked 2026-10-31) |
Where to Buy & Current Price
Prices and promotions change — check the live shop page. Price checked: 2026-10-31 on shop.olares.com (single configuration).
- PromptQuorum earns nothing from any purchase: the links on this page are plain reference links and carry no affiliate tag.
- Confirm the final price, shipping, import duties and warranty terms on the shop page before ordering — the Olares product page links to shipping and warranty policies but does not state them.
Signal | Guidance |
|---|---|
| Official shop price | $4,999 USD (shop.olares.com, checked 2026-10-31) |
| Olares email quote | $4,959 USD, $40 off, shipping in about 2 business days from US and EU warehouses (vendor claim, not verified by PromptQuorum) |
| Olares Black Friday event | 2026-11-17 00:00 to 2026-12-03 00:00 (GMT-8), an extra $300 USD off each unit (vendor claim); order through the shop or email Olares and reference "PromptQuorum Partner Quote" for early allocation and multi-unit bundles |
| Black Friday price (arithmetic) | $4,659 USD per unit if the $300 reduction applies to the $4,959 quote (PromptQuorum's calculation from Olares' email; vendor claim, not verified) |
| Third-party retailers | Not checked |
Alternatives to Consider
The right alternative depends on whether you value size, VRAM per dollar or memory capacity.
- Want the most VRAM per dollar and accept a tower → a desktop GPU build; see the GPU buying guide and the workstation build guide.
- Want large models in one memory pool → an Apple Silicon Mac with large unified memory; see best Mac for local AI.
- Want a smaller, cheaper box for light models → a Ryzen or Intel mini PC; see ASUS mini PCs for local AI and the mini PC roundup for Home Assistant.
- Wondering what fits on 24 GB → the 70B models on consumer hardware guide, which includes the Olares One.
How We Evaluate
PromptQuorum has not tested an Olares One. This page separates three kinds of information so you can see what is confirmed.
- Manufacturer-published: specifications and noise figures from olares.com/one, and the price from the official shop (both checked 2026-10-31).
- Vendor claims: figures Olares sent us by email — shown as claims, never as results.
- PromptQuorum estimate: model-fit and speed reasoning from VRAM size and memory bandwidth, not a measurement.
Frequently Asked Questions
What is the Olares One?
A compact desktop AI computer from Olares with an Intel Core Ultra 9 275HX, an RTX 5090 Mobile with 24 GB GDDR7, 96 GB DDR5-5600 and a 2 TB NVMe SSD, running Olares OS.
How much does the Olares One cost?
The official Olares shop lists $4,999 USD for its single configuration (checked 2026-10-31). Olares also quoted $4,959 and a Black Friday reduction of $300 by email; PromptQuorum has not verified those offers.
Can the Olares One run 70B models?
Yes, with partial offload. A Q4_K_M 70B model is larger than the 24 GB of VRAM, so part of it runs from DDR5 system RAM. PromptQuorum's estimate is about 2–4 tokens/s; Olares emailed 18–22 tokens/s, which is not published on its product page and which we could not verify.
How loud is the Olares One?
Olares publishes 19 dB at idle and 38.8 dB at full power, measured at 25 °C and 0.5 m. An email from Olares quoted under 32 dB at full load, which conflicts with the product page.
Is this an affiliate page?
No. PromptQuorum has no affiliate agreement with Olares, adds no affiliate parameters to links and earns no commission or fee from this page.
Did PromptQuorum test the Olares One?
No. This page uses manufacturer-published specifications, clearly labelled vendor claims and PromptQuorum estimates.