Happy Year of the Horse! Please note that following model update and maintenance date.

GLM-5 on SCRP-Chat

We are proud to offer experimental access to the most intelligent open model to date—Z.ai’s GLM-5. GLM-5 matches the performance of leading commercial models at the moment of its release, only to be superseded by the latest commercial models released immediately after.

Note that the most intelligent model is not necessarily the best model for your task. GLM-5 runs much slower than GLM-4.7, due to a combination of larger model size, unoptimized backend and the model’s tendency to think longer. In a video, we demonstrate how long it takes GLM-5 to solve a simple Math Olympiad problem, in comparison to GLM-4.7 and GPT-OSS-120B.

TL;DR:

  • GLM-5: 120s
  • GLM-4.7: 15s
  • GPT-OSS-120B: 5s

Given the huge speed difference, it only make sense to use GLM-5 if the simpler models cannot handle your task.

As GLM-5 takes a significant number of GPUs to run, we are offering it on an experimental basis so that you can assess its capabilities, before we decide whether to make it a long-term offering.

Maintenance Feb. 20

Two maintenance tasks will be carried out on Friday Feb. 20:

  1. The distributed storage system (~/large-data) needs to be rebooted, in order to fix a problem with its quota system. The system will be offline between 9am-5pm.
  2. The interface of SCRP-Chat, Open WebUI, will be updated to the newest version. This should fix SCRP-Chat’s image editing feature, which was broken in the last update. SCRP-Chat will be offline between 11pm to midnight.