GLM-5 (744B-A40B MoE) Architecture Parameters, Benchmarks & VRAM
744B total / 40B active frontier MIT-licensed MoE built for multi-GPU enterprise clusters.
🧬 GLM-5 (744B-A40B MoE) Architecture & Tensor Specifications
Verified directly against HuggingFace (zai-org/GLM-5) config.json & Model Card
1. Backbone & Dimensions
glm_moe_dsa- Total / Active Params
- 744B / 40B
- Hidden Size (d_model)
- 6,144
- FFN Intermediate Size
- 12,288
- MoE Expert Dim
- 2,048
- MoE Experts
- 256 Routed · Top-8 + 1 Shared
- Activation / Norm
- SiLU (SwiGLU) · ε=1e-5
- Native Tensor Dtype
- BF16
2. Attention & Layer Layout
MLA- Num Hidden Layers
- 78 Layers
- Hidden Layer Layout
- 78-Layer GlmMoeDsa Flagship MoE (256 Routed + 1 Shared Expert, 8 Active)
- Attention Heads (Q / KV)
- 64 Q / 64 KV
- Head Dimension
- 64
- Linear / KV Compression
- DSA Low-Rank KV Compression: kv_lora_rank=512 + qk_rope_head_dim=64
- Vision Encoder / MTP
- MTP 1L
3. Context, RoPE & Tokenizer
Tokenizer- Native Context Length
- 202,752 tokens
- RoPE Configuration
- RoPE (θ=1,000,000 · 原生 202,752 Context)
- Vocab Size
- 154,880 (Untied)
- Special Token IDs
- EOS: [154820,154827,154829] · PAD: 154820
📋 GLM-5 (744B-A40B MoE) Official API & System Card Specifications
Verified against official Zhipu AI developer documentation & pricing
1. Identity & Capabilities
API Spec- API Model ID
- glm-5
- Release Date
- 2026-02-12
- Weight License
- Open-Weight (API Spec)
- Native Reasoning
- ✅ Supported
- Function Calling & JSON Schema
- ✅ 原生支持
- Batch API 异步批处理
- ✅ Supported (50% off)
2. Context, Output & Modalities
Limits- Max Context Window
- 1,024,000 tokens (1024K)
- Max Output Tokens
- 131,000 tokens
- Input Modalities
- Text
- Output Modalities
- Text
- Direct China Access
- ✅ Direct CN + RMB
- Deployment Mode
- Open-Weight + Cloud API
3. Official Pricing & Cache
CNY / 1M- Input / 1M Tokens
- ¥ 4
- Cached Input / 1M
- ¥ 1 -75%
- Output / 1M Tokens
- ¥ 18
- Tiered Pricing Notes
- Zhipu BigModel: Official Zhipu BigModel API
🏆 Official Benchmark Results
Sourced from official System Card / Model Card & technical reports
PhD-level Google-proof science reasoning questions
Humanity's Last Exam evaluated with tool use
A verified subset of 500 software engineering problems from real GitHub issues, validated by human annotators for evaluating language models' ability to resolve real-world coding issues by generating patches for Python codebases.
A multilingual benchmark for issue resolving in software engineering that covers Java, TypeScript, JavaScript, Go, Rust, C, and C++.
Dual-control conversational agent & tool-use benchmark
MCP Atlas is a benchmark for evaluating AI models on scaled tool use capabilities, measuring how well models can coordinate and utilize multiple tools across complex multi-step tasks.
🎛️ Official Sampling & Reasoning Parameters
Recommended generation settings & API parameter constraints
📊 GLM-5 (744B-A40B MoE) Full Benchmark Scorecard
5 public benchmarks · ranks among 154 models · synced 2026-11-20
💻 Coding 3 items
A verified subset of 500 software engineering problems from real GitHub issues, validated by human annotators for evaluating language models' ability to resolve real-world coding issues by generating patches for Python codebases.
MCP Atlas is a benchmark for evaluating AI models on scaled tool use capabilities, measuring how well models can coordinate and utilize multiple tools across complex multi-step tasks.
Terminal-Bench 2.0 is an updated benchmark for testing AI agents' tool use ability to operate a computer via terminal.
🤖 Agents & Tools 2 items
t2-bench is a benchmark for evaluating agentic tool use capabilities, measuring how well models can select, sequence, and utilize tools to solve complex tasks. It tests autonomous planning and execution in multi-step scenarios.
BrowseComp is a benchmark comprising 1,266 questions that challenge AI agents to persistently navigate the internet in search of hard-to-find, entangled information.
Source: ZeroEval / llm-stats.com (✓ = third-party verified, others vendor-reported).
⚡ GLM-5 (744B-A40B MoE) Interactive VRAM & Token Speed Calculator
Physical 96 GB · Load 491.1%
Exceeds usable VRAM by 367.91 GB
MLA compresses multi-batch KV footprint
OLLAMA_CONTEXT_LENGTH=16384 ollama run glm5:744bInsufficient Local VRAM? Ways to Run GLM-5 (744B-A40B MoE)
Rent an hourly high-VRAM cloud GPU pod, or check the 12-month Buy vs Cloud vs API break-even curve.
📊 GLM-5 (744B-A40B MoE) Quantization VRAM Breakdown (16K Context)
| Quantization | Bits/Weight | Weights | KV Cache (16K) | Total VRAM | Quality Note |
|---|---|---|---|---|---|
| FP16 / BF16 | 16 bpw | 1488 GB | 1.47 GB | 1519.68 GB | Lossless Full Precision |
| FP8 / Q8_0 | 8.5 bpw | 790.5 GB | 1.47 GB | 808.23 GB | 99.9% Near-Lossless |
| GGUF Q6_K | 6.56 bpw | 610.08 GB | 1.47 GB | 624.2 GB | Ultra High Quality |
| GGUF Q5_K_M | 5.69 bpw | 529.17 GB | 1.47 GB | 541.68 GB | High Balanced Quality |
| Q4_K_M / AWQ | 4.85 bpw | 451.05 GB | 1.47 GB | 461.99 GB | Sweet Spot (Recommended) |
| GGUF IQ4_XS | 4.25 bpw | 395.25 GB | 1.47 GB | 405.08 GB | Compact 4-bit |
| GGUF Q3_K_M | 3.91 bpw | 363.63 GB | 1.47 GB | 372.82 GB | Moderate Perplexity Loss |
| GGUF Q2_K | 3.15 bpw | 292.95 GB | 1.47 GB | 300.73 GB | Extreme Compression |
🖥️ GPU & Apple Silicon Compatibility for GLM-5 (744B-A40B MoE) (Q4_K_M)
Frequently Asked Questions
What are the exact architecture parameters of GLM-5 (744B-A40B MoE)?
GLM-5 (744B-A40B MoE) uses GlmMoeDsaForCausalLM with 78 layers (78-Layer GlmMoeDsa Flagship MoE (256 Routed + 1 Shared Expert, 8 Active)), hidden_size=6144, 64 Q heads / 64 KV heads (head_dim=64), and vocab_size=154,880.
How much VRAM does GLM-5 (744B-A40B MoE) require?
At 4-bit quantization (Q4_K_M, 16K context), GLM-5 (744B-A40B MoE) requires 461.99 GB VRAM. Full FP16 requires 1519.68 GB.