GLM-4.7 Flash
The strongest model in the 30B class. Balances performance and efficiency for lightweight local deployment. New option for the 24GB-VRAM sweet spot.
Best for strong 30B coder that runs on a single 3090/4090
Sizes 30B
Context 128K
License MIT
Min VRAM (default size, Q4) 18 GB
Rec VRAM 24 GB
What is GLM-4.7 Flash?
GLM-4.7 Flash is Z.ai's open-weights language model. Released in 2026-06, it ships in 1 size (30B) and is licensed MIT. The most popular use case is strong 30B coder that runs on a single 3090/4090.
VRAM and hardware
The 30B size fits in 18 GB of VRAM.
| Size | Min VRAM (Q4_K_M) | Recommended VRAM | Tokens/sec on 3090 |
|---|---|---|---|
| 30B | 18 GB | 27 GB | ~18 tok/s |
How to run GLM-4.7 Flash locally
Option 1: Ollama (simplest)
ollama pull glm-4-7-flash
ollama run glm-4-7-flash Option 2: Mullama (production)
mullama pull glm-4-7-flash
mullama run glm-4-7-flash Option 3: llama.cpp (CLI)
# Download a GGUF from Hugging Face (search "glm-4-7-flash gguf")
./llama-cli -m glm-4-7-flash.Q4_K_M.gguf -p "Hello, AI!" Option 4: Python with Mullama or llama-cpp-python
from mullama import Model, Context
model = Model.load("glm-4-7-flash.Q4_K_M.gguf", n_gpu_layers=99)
ctx = Context(model, n_ctx=4096)
print(ctx.generate("Hello, AI!", 256)) What you can build with GLM-4.7 Flash
- coding
- agentic
- chat
- tools
Hardware it fits on
- Apple Silicon (8GB): 30B in MLX
- Apple Silicon (16GB): 30B in MLX
- Apple Silicon (32GB): 30B in MLX or GGUF
- Apple Silicon (64GB+): 30B in MLX
- RTX 3090 (24GB): 30B at Q4_K_M
- RTX 4090 (24GB): same as 3090, ~30% faster
- RTX 5090 (32GB): 30B at Q8_0
- 2× RTX 6000 Ada (96GB total): 30B at Q4_K_M
Related models in the GLM family
- GLM-5 — 744B (40B active MoE) · frontier-tier reasoning and agentic engineering
- GLM-5.1 — 744B (40B active MoE) · next-generation frontier reasoning, agentic engineering
- GLM-5.2 — 744B (40B active MoE) · latest Z.ai frontier reasoning model (2026 mid-year)
- GLM-4.7 — mid-size MoE · advanced coding and agentic workflows
Other models you might consider
- Llama 3.1 (Meta) — 8B, general-purpose chat, agentic workflows, code
- Llama 3.2 (Meta) — 1B, low-end hardware, mobile, edge deployment
- Llama 3.2 Vision (Meta) — 11B, vision + chat on desktop GPUs
- Llama 3.3 (Meta) — 70B, best-in-class 70B for desktops
- Llama 4 (Meta) — Scout 17B (109B MoE), vision + long context (10M tokens)
- Qwen 2.5 (Alibaba) — 0.5B, strongest 7B-32B on consumer hardware