Model
Model explorer

GLM-Z1-32B-0414

OPEN
Zhipu AI · GLM-Z1 family · released Apr 14, 2025

Open reasoning model built on GLM-4-32B-0414 via cold-start plus extended RL; evaluated on AIME24/25, MATH-500, GPQA and LiveCodeBench, marketed as much faster than DeepSeek-R1 (vendor claims up to 200 tok/s vs 671B R1).

ReasoningCodingVisionFunction callingTool useAgentic
1421.3
Elo · rank #135
Parameters
32B
Active params
32B (dense)
Context
32K tokens
Architecture
Dense transformer (reasoning-tuned GLM-4-32B-0414 via cold-start SFT + RL)
License
MIT
Languages
2+
API price (in/out)
No hosted API
Modalities
text
Benchmark results
Bar shows position within the tracked field; marker = field best
AIMEMath80.8%#73
best: GPT-5.2 · 100.0%
Arena-HardHuman preference90.6%#11
best: Qwen3-235B-A22B (Non-Thinking) · 96.1%
BFCL v3Agents70.5%#7
best: Hunyuan-A13B · 78.3%
GPQA DiamondReasoning66.1%#137
best: GPT-5.6 · 94.6%
IFEvalReasoning84.5%#67
best: Gemma 4 26B A4B · 98.5%
LiveCodeBenchCoding59.1%#99
best: DeepSeek-V4-Pro (Think Max) · 93.5%
OmniMathMath68.4%#3
best: Phi-4-reasoning · 76.6%
SuperGPQAReasoning52.6%#20
best: Qwen3.7-Max · 73.6%
Run it locally
VRAM @ Q4
20 GB
VRAM @ FP16
64 GB
Fits on (Q4)
RTX 3090 24GBRTX 4090 24GBRTX 5090 32GBM4 Pro 48GBM3 Max 128GBM3 Ultra 512GBA100 80GBH100 80GBH200 141GBB200 192GB
Vendor claims ~200 tok/s on Zhipu's own MaaS platform; not an independently measured RTX 4090 llama.cpp benchmark.
Quantizations
GGUF Q4 · GGUF Q8 · MLX · GPTQ
GLM-Z1 family
Elo progression across releases
GLM-Z1-32B-0414Apr 20251421.3
Resources
HFzai-org/GLM-Z1-32B-0414GHzai-org/GLM-4APIAvailable via Zhipu AI API
API price weights · each benchmark row carries its own source badge (see methodology)