Model
Model explorer

Qwen2.5-Math-72B

OPEN
Alibaba · Qwen2.5-Math family · released Sep 18, 2024

Base math-specialized model continual-pretrained from Qwen2.5-72B on Qwen Math Corpus v2; Instruct sibling adds CoT+TIR reasoning.

ReasoningCodingVisionFunction callingTool useAgentic
919.6
Elo · rank #231
Parameters
72.7B
Active params
72.7B (dense)
Context
4K tokens
Architecture
Dense transformer, 80 layers (GQA 64Q/8KV)
License
Qwen (Tongyi Qianwen License)
Languages
2+
API price (in/out)
No hosted API
Modalities
text
Benchmark results
Bar shows position within the tracked field; marker = field best
GSM8KMath90.8%#39
best: Llama 3.1 405B · 96.8%
MATH-500Math66.8%#104
best: GPT-5 · 99.4%
MMLU (STEM)Knowledge82.8%#2
best: Falcon-H1 34B · 83.6%
Run it locally
VRAM @ Q4
47 GB
VRAM @ FP16
145 GB
Fits on (Q4)
M3 Max 128GBM3 Ultra 512GBA100 80GBH100 80GBH200 141GBB200 192GB
Q4 weights (~47GB) exceed a single RTX 4090's 24GB VRAM; requires multi-GPU or CPU offload for local inference.
Quantizations
GGUF · AWQ · GPTQ Int4 · MLX
Qwen2.5-Math family
Elo progression across releases
Qwen2.5-Math-72BSep 2024919.6
Resources
HFQwen/Qwen2.5-Math-72BGHQwenLM/Qwen2.5-MathDOCSqwenlm.github.io/blog/qwen2.5-math
API price weights · each benchmark row carries its own source badge (see methodology)