[hnlmorg]

> Open models like GLM 5.2 are getting good enough to handle 90% of tasks, and will eat most of their usage unless they start serving it at cost.

Most people either don’t have the money to buy hardware to run open models and/or don’t have the utilisation to make renting cost effective.

[m_ke]

https://openrouter.ai/models

you can use these in hermes, cursor, openclaw, opencode, etc with 2 lines of config that claude code will happily do for you if you ask

GLM 5.2, deepseek 4 Flash and the newly released Hy3 are Opus 4.8 and Sonnet 4.6 level models at a tiny fraction of the cost.

I'm on the $200 claude plan and blew through my weekly limits with Fable in a day, then ended up wasting $20 with opus 4.8 overages in an hour to finish out work in active sessions. Since then I've been using GLM 5.2 with openrouter + opencode and am spending less than $5/day for equivalent output.

[hnlmorg]

Ahhh that makes a lot more sense. Thanks for the detailed explanation.

[Systemerror7A69]

GLM 5.2 is monstrous in size, no one is running that on their own hardware.

But it's a very competitively priced model other providers can offer (since it's open) so it's a much cheaper alternative than claude in practice.

I assume that is what they meant.

[Aeolos]

Fortunately, multiple model providers are offering open weights models at significantly lower prices than closed models.

[NetOpWibby]

This is where I’m at. While I’d LOVE to have a powerful near-frontier model at home, I don’t have the extra funds necessary to purchase a tricked-out Mac Studio…and if I did, I’d pay off debts.

[datakan]

Probably wouldn't matter if you had the funds either since MacStudios are 3-4 months out for purchase.

[netruk44]

And also the weights for GLM 5.2 are 1.5 TB. The only Studios Apple sells today are hard-capped to 64 GB (M4 Max) or 96 GB (M3 Ultra) only.

You'd need at least 24 new M4 Max Studios, or 16 new M3 Ultra Studios, or 3 used 512 GB M3 Ultra Studios just to power one GLM 5.2 instance. And even then, you're probably looking at < 5 tokens per second.

Personally, I think it makes way more sense to pay a model provider $3/1M tokens.

[m_ke]

Deep Learning models are designed to get max throughput on GPUs, which ends up being batched workloads.

You'll never get proper price competitive utilization on personal hardware vs a cloud inference provider that can batch and pipeline requests optimally to maximize utilization, unless you yourself start running batch jobs.

Even once local hardware and models catch up to todays frontiers, by that time there will be 10x better cluster sized models available at a similar discount.