Alibaba Qwen2.5-72B-Instruct (Q4 quant)

Name: Alibaba Qwen2.5-72B-Instruct (Q4 quant)
Brand: Alibaba

Open weights. Your GPU, your rules.

Stylized line drawing of the Alibaba Qwen2.5-72B-Instruct (Q4 quant)

Qwen2.5-72B-Instruct, Qwen's 72B-class open-weights instruct model (September 2024). Qwen's own Q4_K_M build is 44 GB, so dual 24 GB cards in tensor-parallel, not a single 32 GB card. Titled Qwen3 72B until 2026-09-24; Qwen3 has no 72B model.

Remix this rig in the arena Compare with another

Specs

parameters: 72B
quant: Q4_K_M
vram required gb: 44
context window: 128K tokens

Buy

Hugging Face ↗
Direct download. No affiliate relationship.

Qwen2.5-72B is the Soul you pick when you don't want an API dependency. Precision ceiling is the main gap versus Opus or GPT-5; Resilience is top of the chart.

Correction, 2026-09-24: this page was titled Qwen3 72B and linked to a Qwen3-72B-Instruct repository that does not exist; Qwen3's dense releases stop at 32B. It now describes Qwen2.5-72B-Instruct, the 72B-class open-weights model Qwen did release.