744B parameters with sparse MoE; activates 40B per token. 200K context, MIT license. Ranks #1 open-weight with frontier performance.
Released February 2026; #1 ranked open-weight model on Artificial Analysis and LMArena Text Arena.
Updated 11 Feb 2026Checked 10 Oct0 updates this week
Router network selects experts per token. Drastically reduces compute vs. dense models.
Outputs 2.76 tokens per step on average; nearly triples throughput.
160B tokens of software engineering data explains strength on SWE-bench.
| Spec | GLM-5 |
|---|---|
| Total parameters | 744 billionR (reported) |
| Active parameters per token | 40 billionR (reported) |
| Context window | 200K tokensR (reported) |
| Training tokens (all stages) | 28.5 trillionR (reported) |
| AI2 Intelligence Index | 50R (reported) |
| SWE-bench Verified score | 77.8%R (reported) |
| AIME 2026 score | 92.7%R (reported) |
| API price (input tokens) | $1.00 per millionR (reported) |
| API price (output tokens) | $3.20 per millionR (reported) |
R reported by the company
Z.ai (Zhipu AI)
GLM-5, GLM-5.2, open-weight models, coding agents
China's first major LLM company to go public. Founded 2019 by Tsinghua professors. Develops GLM open-weight models. Debuted Hong Kong Stock Exchange January 2026 at $7.4B, raising $558M.
