GLM 5
zai-org · released 2026-02-11 · mit license
GLM-5 is Z.ai's 744B-parameter (40B active) MoE frontier model (Feb 2026, MIT), ranked #1 open-weight on the Artificial Analysis Intelligence Index, with 77.8% SWE-bench Verified, 92.7% AIME 2026 and 86.0% GPQA Diamond (vendor).
Key specs
| Type | Local open-weight |
|---|---|
| Parameters | 753.86B total · MoE, — active |
| Architecture | glm_moe_dsa |
| Context window | 203K tokens |
| Knowledge cutoff | — |
| Modalities | text |
| Recommended backends | — |
| Minimum viable rig | — |
Benchmark scores
| GPQA Diamond | 86% |
|---|---|
| SWE-bench Verified | 77.8% |
| AIME | 92.7% |
| MMLU-Pro | — |
| BFCL v3 (tool use) | — |
| Composite score | 8.38 |
| Community rating | No reviews yet |
VRAM & disk per quantization
| Quant | VRAM | Disk | RAM | Context |
|---|---|---|---|---|
| Q4_K_M | 438.7 GB | 437.2 GB | — | 203K |
Strengths & weaknesses
Strengths: #1 open-weight on the Artificial Analysis Intelligence Index (50) at release; Strong agentic/coding and long-horizon planning; Permissive MIT license
Weaknesses: Text-only, no native multimodal; Verbose and relatively expensive to run