GLM-4.7-Flash (30B-A3B)
30B-A3B MoE, MIT license. Still a community favorite for 16-24 GB setups months after release.
Recommended quants
- Unsloth UD-Q4_K_XL (17.5 GB)
unsloth/GLM-4.7-Flash-GGUF - UD-Q3_K_XL (13.8 GB) / UD-Q2_K_XL (11.9 GB)
unsloth/GLM-4.7-Flash-GGUF - MLX 6/8-bit
lmstudio-community/GLM-4.7-Flash-MLX-* - AWQ
QuantTrio/GLM-4.7-Flash-AWQvLLM.
Chat template Unverified
Template behaviour not checked for this card.
Fix: Use the template shipped with the official repo / quant.
Sampling settings
| Default (Z.ai eval settings) | temp 1.0, top_p 0.95 |
|---|---|
| Other Z.ai eval settings | temp 0.7 and temp 0 used for some benchmarks |
Runtimes
| vLLM | Works | --tool-call-parser glm47 --reasoning-parser glm45 (official). |
|---|---|---|
| llama.cpp | Works | Unsloth GGUFs. |
| Ollama | Unverified | Library page glm-4.7-flash exists. |
| LM Studio | Works | lmstudio-community uploads. |
| MLX | Works | lmstudio-community MLX quants. |
Tool calling in agent tools
| Claude Code (local endpoint) | Unverified | |
|---|---|---|
| OpenCode | Unverified | |
| Cline | Unverified | |
| Aider | Unverified |
Known issues
Sources checked
Spotted something wrong or a new fix? Corrections are reviewed before publishing. (Submission form coming soon.)