Free big models, whoever picks them up at will, and whoever adjusts them will limit them

The last time a cross-section review of free large models was posted, the most frequently asked question in the comments was: How much can these free ones be collected? Will it be a limit if I adjust it twice?

Good question. The horizontal review talks about "which one can use for free", not "how hard it can be used." This article will make up for this. I did something quite damaging today: I sent a bunch of requests to each family for free models without breathing, to see who could let me pick them up at will, and who would turn against me the second time. Data from today's real run.

Let's start with an anti-common sense point: how much can you collect for free? In fact, the two numbers are fighting.

One is the daily amount, how much is the total amount you will be given in one day. The other is to continuously limit current, so you can adjust several times in a short period of time. These two often do not match: some people give a generous daily quota, with millions of tokens a day, but if you adjust it a little faster, the second request will hit you with 429 in the face. Therefore,"a large free quota" and "being able to collect at will" are two different things.

I took six free large models and sent 15 requests in a row without interval, but the results were ridiculously bad.

!

Let's see what you can pick up at will first. Wisdom's GLM-4-Flash, there is no limit to one in 15 shots. It has always been claimed to be permanently free, uncapped, and anti-concurrency. This time, the actual measurement has been confirmed. It is the king of wool collection. You can adjust it at will without frowning. Kimi K3 is also in the 15th generation. This is the latest and largest open source model in this wave. Now it has no current restriction and can be used; but I guess it has just been released and has not had time to tighten it yet, so I want to be used as soon as possible. Groq is also available in fifteen, and there is no limit to continuous bursts, but it has a ceiling for total volume, 100,000 tokens a day, and stops when the amount is collected.

Just one more note and you'll turn your face off. There is also a GLM-4.7-Flash, which is half a version number short of the 4-Flash above. The treatment is vastly different: the second one in a row is 429, and 14 out of 15 are limited, so it is basically impossible to connect. ModelScope's DeepSeek-V4-Flash is even more ruthless. The first of fifteen requests is limited, and all of them are wiped out. Today, there is not a drop.

The most interesting is Cerebras. it is very generous in daily quota, giving you a million tokens a day, listen, you can make whatever you want. however, i kept repeating it, the second one was 429, and only three out of fifteen succeeded. This is the typical example of "a large amount but you can't collect it": if the total amount is given, but the QPS card will die, you can only adjust it slowly one by one. If you want to use multiple threads to collect it, you can directly restrict it.

!

So I really want to stabilize white prostitutes and give you some tricks that I have actually measured.

The main player releases GLM-4-Flash. It has no caps or limits on consecutive bursts, and it is the only thing that allows you to open it up. Although it is not the smartest, the advantage is that it will not be annoying no matter how you adjust it. It is the most reliable to do high-frequency and batch work.

If you want to get a stronger one, go to Kimi K3 as soon as possible. It now has no current restriction and is the latest big model, so it can get good things; but the free window can be closed at the request of you, and don't expect it to remain so generous.

Those with a large daily quota, like Cerebras, don't think about raking it out. The total amount is large, but you have to leave some space between each request, not concurrent, and come one by one honestly to collect the one million.

As for Smart Spectrum 4.7-Flash and ModelScope, which are limited in continuous bursts, just be a spare tire. Don't expect to rely on it to do serious work.

There are many things you can gain from free models, but each family has different temperaments. Some of them can be collected at your disposal, and some give you a look. Find out who can open it, who can only sip it, and who simply don't touch it. Only then can you collect more without worrying.