Architecture
data/models.yaml. Every label is auditable
against the model's sources.
Specs
- Architecture
- moe
- Total params
- 355B
- Active params
- 32B
- Context window
- 128K tokens
- Attention
- unknown
- Position encoding
- unknown
- Pretraining tokens
- 23.0T
- Post-training
- sft, rlhf
- OSI-approved
- yes
- Data released
- no
- Training code
- not released
Benchmarks
Each score carries the date it was published; we never infer or interpolate missing scores.
Code
| SWE-Bench Verified | 64.2 | as of 2025-07-28 | source ↗ |
Math
| AIME 2024 | 91.0 | as of 2025-07-28 | source ↗ |
Available quantizations
GGUF llama.cpp's container; the common local format, k-quants from Q2 to Q8.
runs on llama.cpp, Ollama
Verified via the Hugging Face model tree ↗. Community quantizations change over time; the families shown are those with published weights at audit time.
Notable innovations
- · MIT-licensed 355B MoE
- · Hybrid thinking / non-thinking modes
- · Multi-Token Prediction layers with speculative decoding
- · 23T-token pretraining
Lineage
First Zhipu ARC flagship; 355B MoE under MIT. New pretrain, architectural break from the dense GLM-4 9B series.
Derivatives