zai-org

/

GLM-4.7

like 1.92k

Follow

Z.ai 9.48k

I have the same experience on yearly PRO plan.

At long last, someone read my email, only to reply with, “Well, you’ve used so many tokens, so what’s the fuss?” I had to explain, I bought it for a whole year, not a bag of magic beans, and it’s barely been a few weeks—so am I supposed to just twiddle my thumbs until next New Year? Then, in a plot twist, they shrank the context length from 250k to 200k, and suddenly things improved… though now the speed plays hide-and-seek depending on the day and time.

I came here after looking for "z.ai slow", I dodged the bullet!!!! The part about "glowing youtube reviews" had my try this, but I did not buy yet, happy to stay with Claude Pro.

To be fair, it’s improved a lot lately. They must have boosted their capacity because it’s much faster now. That said, plenty of new players have entered the market too, like Moonshot.AI and others.

You get what you pay for, I guess. I'm using it as a supplement to my Claude Max subscription, so I can't offer in-depth feedback on performance. That said, I think this discussion is just off-topic — the focus should be on the published model itself, not the hosting.

For what it's worth, anyone can run it locally. Rent a decent GPU for a few hours, download the weights, and you're set — for as long as you can afford the compute. Which, spoiler alert, isn't very long. Welcome to the obvious club.

Now imagine your entire business model is built on overbooking capacity. Selling more seats than the plane can hold works great — until everyone shows up.

Oh, and for those of you in the EU: consumer protection law is on your side. If the service doesn't deliver what was advertised, you're entitled to a refund. Sometimes the most productive feedback is the kind that hits the balance sheet.

I tried Z.Ai for a while and was blown away by its lightning-fast reasoning. The glowing YouTube reviews made it sound like the best thing since sliced bread, so I jumped in and grabbed a Christmas discount on the yearly subscription. Naturally, the moment I upgraded, Z.Ai disappeared into the digital void. Now it’s slower than a snail on vacation, and support—only reachable by email—has completely ghosted me. I’m about ready to call my credit card company to report a possible scam. Moral of the story: if you like it, stick to the monthly plan. Once they’ve got your yearly payment, they have zero reason to impress you—they’re already off spending your money on virtual piña coladas. Considering emails are the only contact option and they always bounce back, it’s disappointing that a supposedly cutting-edge AI company can’t even handle basic communication.

I just reached out to them and requested, according to EU regulations, a GDPR data request. The deadline for the data request is one month. If I forget to report the results, please let me know in this discussion.

A good model is only as good as its processing speed. Whats the point of a great model but it takes forever to think or return any results? That being said, they have certainly improved a lot in terms of processing speed. Though occasionally, I do get a "API error", where I have to cancel the session and continue again. Not sure if its a quick of Kilo Code or what. But its useable now and I use it as a supplement to my Claude and Antigravity as well.

Sadly, It's a very slow model. I have been a subscriber for a while. It's cheap, but you can't get usage out of it like with other services.

I use it with Ralphy alot and just leave it running for the whole day. But it takes 8 hours to do some pretty basic stuff.

It's not a bad model, GLM 5 looks promising - but I'd say the benchmarks are cherry picked more than with other models.