Beelink SER8 review: the $520 box that got me off ChatGPT

Some links below are affiliate links. How this works.

The Beelink SER8 is a mini PC about the size of a hardback book. It costs around $520 with 32 GB of memory and a 1 TB drive, and it has been sitting under my monitor since March answering questions, cleaning up my writing and searching a folder of personal notes I would never upload to a company server.

It is not as sharp as the paid AI services. But it is close enough, often enough, that I let my subscription lapse and did not miss it.

What it’s actually like to live with

You do have to set it up. This is the part nobody selling you on “local AI” mentions. You install a small program — I use one called Ollama — download a model file the way you’d download a big game, and point a chat window at it. Start to finish it took me about twenty minutes, most of it waiting for the download.

After that it just runs. I ask it things from my laptop and my phone, both on the home wi-fi, and it answers in about two seconds. No login, no “you’ve hit your limit,” no wondering who’s reading it. When the power blipped in a storm it came back on its own.

The model I run day to day is a mid-size one — capable enough to draft an email, summarize a long document, or talk through a plan. When I ask it to do something genuinely hard, like reason through a tricky bit of code, you can feel it straining. That’s the ceiling of this machine, and it’s worth being honest about.

Where it’s good

For text, it’s better than I expected. The 780M graphics chip built into the processor isn’t fast, but it shares the machine’s 32 GB of regular memory, and for AI work having enough memory matters more than raw speed. A mid-size model loads in, stays loaded, and responds quickly enough that I stopped thinking about it.

Idle power is about 11 watts. Leaving it on around the clock costs me somewhere around $13 a year. That’s the number that makes an always-available home AI actually make sense.

Where it bites

The largest, smartest models — the ones that get within shouting distance of the big services — need more memory bandwidth than this chip has. You can technically load a big model; you won’t enjoy waiting for it.

Image generation works but is slow enough that you won’t do it twice. If making pictures is your reason for wanting local AI, this isn’t the machine — you want something with a real graphics card.

And the fan spins up under load. It’s not loud, but in a quiet room you’ll know the model is thinking.

Who should buy it — and who shouldn’t

Buy it if you mostly want a private, always-on assistant for text: writing, summarizing, answering questions, searching your own documents. Get the 32 GB version and know you can add memory later.

Buy something else if you want to run the biggest models (look at a 128 GB machine) or you care about image generation (you want a graphics card). And if you only use AI a few times a week, honestly, a subscription is cheaper and less hassle — the SER8 pays off when it becomes part of your day.

The specs, in plain English