GLM 4.6
zai-org · released 2025-09-30 · mit license
GLM-4.6 is Z.ai's 357B-parameter (32B active) MoE (Sept 2025, MIT) with a 200K context; it scores 30 on the Artificial Analysis Intelligence Index and reports 82.8% on LiveCodeBench v6. A leading open-source coding model at launch.
Key specs
| Type | Local open-weight |
|---|---|
| Parameters | 356.79B total · MoE, — active |
| Architecture | glm4_moe |
| Context window | 203K tokens |
| Knowledge cutoff | 2025-03-31 |
| Modalities | text |
| Recommended backends | — |
| Minimum viable rig | — |
Benchmark scores
| GPQA Diamond | — |
|---|---|
| SWE-bench Verified | — |
| AIME | — |
| MMLU-Pro | — |
| BFCL v3 (tool use) | — |
| Composite score | — |
| Community rating | No reviews yet |
VRAM & disk per quantization
| Quant | VRAM | Disk | RAM | Context |
|---|---|---|---|---|
| Q4_K_M | 208.4 GB | 206.9 GB | — | 203K |
Strengths & weaknesses
Strengths: Strong real-world coding in agent frameworks (Claude Code, Cline, Roo); 200K context, ~30% better token efficiency than GLM-4.5; MIT license
Weaknesses: Trails Claude Sonnet 4.5 in pure coding; Text-only