Qwen3-72B 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.
Alibaba · cloud-model
Verified 1mo ago
Alibaba Qwen3-72B-Instruct (Q4 quant)
Frontier-adjacent. Open weights. Your GPU, your rules.
The best open-weights Soul in its class. Needs a 32 GB Heart to sit comfortable at Q4; fits on dual 24 GB in tensor-parallel.
Specs
- parameters
- 72B
- quant
- Q4_K_M
- vram required gb
- 44
- context window
- 128K tokens
Buy
- Hugging Face ↗
Direct download. No affiliate relationship.