GLM-5.1
Active Parameters
754B
Context Length
200K
Modality
Text
Architecture
Mixture of Experts
KV Cache Rate
88 KB/tok
License
MIT
Release Date
7 Apr 2026
Knowledge Cutoff
-
API Pricing (per 1M)
Input: $1.40 · Output: $4.40
Hardware Sizing
Precision
Context
Workload Memory: 1491.19 GB
FP16 · 8K context
Base Weights: 1478.94 GB
KV Cache: 9.75 GB
Runtime & Graphs: 2.50 GB
Minimum Required Hardware
Cluster: 16x NVIDIA H200 141GB
1491.2 / 2030.4 GB usable · TP 8 · PP 2
Determine requirements for datacenter hardware:
Upgrade to Pro
Architecture Diagram
Evaluation Benchmarks
Rank
#53
| Benchmark | Score | Rank |
|---|---|---|
Graduate-Level QA | 0.862 | 40 |
Agent Arena | -2.22 | 44 |
General Text | 1465 | 47 |
Web Development | 1508 | 56 |
Agentic Index | auto 0.24 | 71 |
Coding Index | 0.56 | 79 |
Intelligence Index | auto 0.26 Standard 0.24 | 101 117 |
Rankings
Overall Rank
#53
Coding Rank
#73
About GLM-5.1
GLM-5.1 is Z.ai's flagship model for long-horizon agentic coding tasks. Built on a novel GlmMoeDSA architecture with 754B total parameters (256 routed + 1 shared experts, 8+1 active per token) across 78 layers, it combines Gated DeltaNet linear attention with standard attention and sparse MoE feed-forward networks — enabling efficient inference while delivering top-tier intelligence. Achieves state-of-the-art 58.4% on SWE-Bench Pro, 63.5% on Terminal-Bench 2.0, 95.3% on AIME 2026, and 86.2% on GPQA-Diamond. Uniquely designed for 8-hour sustained autonomous execution — breaking complex engineering tasks into iterative experiment-analyze-optimize loops. Supports 200K context window and 128K max output tokens. Available via API as glm-5.1 on Z.ai and BigModel.cn. Released April 7, 2026 under MIT license.
Technical Specifications
Attention
Attention Structure
Multi-Layer Attention
Attention Heads
64
Key-Value Heads
64
Attention Head Dimension
64
Position Embedding
ROPE
RoPE Theta
1,000,000
Normalization
RMS Normalization
Activation Function
SwigLU
Dimensions
Hidden Dimension Size
6,144
Number of Layers
78
FFN Intermediate Size (Dense)
2,048
Multi-Token Prediction Heads
1
Tokenizer
Vocabulary Size
154,880
Mixture of Experts
Total Expert Parameters
40.0B
Number of Experts
257
Active Experts
9
Shared Experts
1
FFN Intermediate Size (per Expert)
2,048
Dense Layers Before MoE
3
About GLM-5.1
GLM-5.1 is Z.ai's next-generation flagship model for agentic engineering, built on a novel hybrid MoE architecture (GlmMoeDSA) combining Gated DeltaNet linear attention layers with standard attention and sparse MoE feed-forward networks. It achieves state-of-the-art performance on SWE-Bench Pro (58.4%) and is designed for long-horizon autonomous tasks, capable of sustained execution for up to 8 hours. With 754B total parameters and a 200K context window, GLM-5.1 delivers strong performance across coding, reasoning, tool use, and agentic benchmarks. Released open-source under the MIT License.
Other GLM-5.1 Models
- No related models available