- likes
- 47
- comments
- 1
Post
GLM-5.3 is now running on SambaCloud at over 200 t/s. It’s Z.ai’s flagship model for software engineering and long-horizon agent work, with state-of-the-art results among open-weight models on Terminal-Bench 3.0 and Agents’ Last Exam, using fewer output tokens per task than several closed frontier models. A typical agent run is 40+ model calls where every step waits on the last. Fewer tokens per step plus 200+ t/s means the whole run finishes while you’re still watching. Works with Claude Code, Codex CLI, OpenCode, and ZCode through a base-URL swap.