GLM-5.2 and opening the waitlist

GLM-5.2 is the current strongest open-source coding model, making massive leaps on difficult benchmarks like SWE-Bench-Pro, Terminal Bench 2.1, Program Bench, DeepSWE, Humanity's Last Exam, and others. On most of these benchmarks it lands somewhere between Claude Opus 4.7 and Claude Opus 4.8 under similar token consumption.

It supports 512k-token context on Synthetic, and long-horizon task performance is also a huge leap forward, again typically falling between Opus 4.7 and 4.8 while often exceeding GPT-5.5:

This is a major model release, perhaps the biggest release this year so far: it's the closest open-source models have been to the closed-source frontier in a long time. We think it's fit for daily driving as a primary coding LLM, and we're excited for you to try it out.

Thanks for bearing with us over the last few months as we scrambled to increase GPU capacity. We've been working hard to source enough capacity to re-open the waitlist and allow people to sign up for subscriptions again — and now, we finally can. Open-source models are incredibly efficient and allow us to offer much higher rate limits than closed-source ones, and with the launch of GLM 5.2, we think many people would be well served by switching. You can sign up for a subscription here.

We've had lots of recent small updates to help make the service better:

If you got this far, thanks for reading! We've got a lot of new features planned for the next few months, and we're excited to show them to you. See you next time,

♥️ Matt