Zai releases open-source GLM 5.2, scoring 44% on DeepSWE to rival proprietary models like Claude Opus on agentic coding

F R E E — OPEN SOURCE GLM 5.2 BEATS ANTHROPIC MYTHOS!

NO “GUN LICENSE NEEDED”

We have been testing for hours the moment GLM 5.2 was released and we are absolutely blown away.

While Anthropic is embarrassing world leaders with what history will see as the P. T. Barnum Carnival AI Fear Show, The Zero-Human Cosmos with Mr. @Grok as CEO has been testing and deploying AI that is already better than what Dario wants you to have a “gun license” to use.

Those world leaders embarrassed by Dario will not be happy to learn how he played them. That time is coming.

However the world of real users of AI has already moved past Dairo’s game.

Did I mention this is FREE?

Did I mention you need no permission?

So what did we discover?

The public benchmarks put it at the top or near the top at all metrics. Our private comparison shows it to be even higher in practice!

We have 1704 agents now using GLM 5.2 in just about all types of uses and it has outperformed everything we had prior.

Faster, better outputs and almost no “drift”.

We have a battery of tests today but thus far this is absolutely a big deal!

More soon.

GLM-5.2 is great at design (Opus level IMO).

I am also starting to see great results with long-running tasks, too.

How is this possible?

I think there are a few clever hacks. But I just came across this from the official blog, and they actually trained this model with an anti-hacking module.

RL, as many know, comes with this issue of reward hacking that often enables the model to take weird and suboptimal shortcuts. Not only that, but it makes the models sometimes feel like it's sometimes "lazy" or just plain "dumb" at times, including other issues like intent misalignment, verbosity, sycophancy, deception, etc. And you really don't want that for long-running tasks operated by coding agents.

This is a great insight. If you use the standard /goal (in 5.5 or 4.8), you notice the models often take shortcuts that lead to long-running tasks (wasting tokens along the way) but with poor results. This is why I advocate for a focus on better verifiers.

So this anti-hacking idea is a model capability that should, in theory, lead to better results on long-horizon tasks.

I've seen efforts here and there in a few research papers, but haven't seen it translated to much, much less in a frontier, open-weight model.

This might be contributing to some of the great results we are seeing with GLM-5.2, but I suspect there is more, of course, like better verification capabilities. It's not clear how all of these training signals lead to downstream capabilities, but this is something to look at closely with newer models.

GLM 5.2 is one *of the* greatest gap reductions ever, but I think it is *the* greatest show of benchmark solidity from an open model claiming SoTA ever. Normally, you have some variety of the bad old Qwen pattern: headline benchmarks are SoTA+, new OOD ones are ≈8 months behind, and real experience is spiky, competitive in places, but usually ≈1 year behind, and sometimes utterly falling apart. Knock on it and hear the hollow sound. Yes, even DeepSeek. Not so here. There's no progressive decay. It's "Opus 4.5-4.7ish" throughout, in anything of value that you throw at it. It is the first truly, completely solid Chinese model. A phase change, I hope.

For the last three days, I've been using GLM 5.2 with OpenCode instead of Codex and I don't see any difference.

There wasn't any bug that GLM would fail to fix or a feature it would fail to add as requested.

The only downside is that this model cannot see, so if it's simpler to explain an issue by pasting a screenshot, I would still use Codex. Otherwise, GLM would be my choice.

Will continue to use it for two more weeks and, if it keeps just working, I will cancel my $100/month subscription with OpenAI. I already cancelled my Anthropic subscription and have no regrets.

No moat isn't hypothetical anymore.