Qwen 3.5 (Withdrawn Listings) — Alibaba Cloud 的本地 AI 模型
作者： Jakub Rusinowski · 最后更新：
Alibaba's most ambitious open-weight model release. Qwen 3.5 (397B) is a natively multimodal model supporting text, vision, and video understanding in a single architecture. Ranks among the top 3 open-weight models globally and competes directly with GPT-4o and Gemini 2.0 Pro on multimodal benchmarks.
变体
Qwen 3.5 (Withdrawn Listings) 最小的变体在 Q4_K_M 下约需 9 GB 显存——量化权重加框架开销，不含 KV 缓存。
| 模型 | Q4 下显存 | 显存 | 上下文 | 运行 |
|---|---|---|---|---|
| Qwen 3.5 14B 14B | ~9.3 GB | 128,000 | ollama run qwen-3-5-legacy | |
| Qwen 3.5 72B 72B | ~44.3 GB | 128,000 | ollama run qwen3.5:72b | |
| Qwen 3.5 397B 397B | ~240.5 GB | 128,000 | ollama run qwen3.5:397b |
显存为 Q4_K_M 下的量化权重加开销，与 GPU 与显存检测器使用同一引擎计算。
如何在本地运行 Qwen 3.5 (Withdrawn Listings)
安装 Ollama，然后拉取标签。
ollama run qwen-3-5-legacy在上方选择一个尺寸，查看它自己的显存、速度估算和安装命令。
许可证
Commercial use permitted. No usage restrictions beyond attribution.
适用于: Qwen 3.5 14B, Qwen 3.5 72B, Qwen 3.5 397B推荐 GPU
目录中能在本地运行 Qwen 3.5 (Withdrawn Listings)（至少 9 GB 显存）的最便宜 GPU 是 Intel Arc B570 (10 GB).
Qwen 3.5 (Withdrawn Listings) — 常见问题
How much VRAM does Qwen 3.5 (Withdrawn Listings) need?
Qwen 3.5 (Withdrawn Listings) needs about 9 GB VRAM at Q4_K_M quantization for its smallest variant. Variants: Qwen 3.5 14B (9 GB, Q4_K_M); Qwen 3.5 72B (44 GB, Q4_K_M); Qwen 3.5 397B (240 GB, Q4_K_M). On Apple Silicon, unified memory counts toward this requirement.
Can I run Qwen 3.5 (Withdrawn Listings) on an RTX 4090 (24 GB)?
Yes — Qwen 3.5 (Withdrawn Listings) runs on an RTX 4090 (24 GB) and other 24 GB cards such as the RTX 3090. Smaller variants also fit comfortably on 8–16 GB GPUs at Q4_K_M.
What quantization should I use for Qwen 3.5 (Withdrawn Listings)?
Q4_K_M is the best balance of quality and VRAM for Qwen 3.5 (Withdrawn Listings) in most cases. Choose Q8_0 for near-lossless quality if you have spare VRAM, or smaller quants (Q3/Q2) only when memory is tight.
How do I run Qwen 3.5 (Withdrawn Listings) with Ollama?
Qwen 3.5 (Withdrawn Listings) has no local Ollama tag — the published tag is cloud-hosted, so running it sends your prompts to a hosted GPU rather than your own machine.