Qwen2 models

15 Qwen2 models from Qwen on Hugging Face, from 494M to 72.7B parameters, published by Qwen in BF16, with contexts from 32K to 128K tokens. The smallest official model, Qwen2-0.5B, needs about 1.1 GB of VRAM at its published precision; the cheapest live fit is RTX 3070 at $0.088/hr.

Part of the Qwen series · Previous generation: Qwen1.5 · Next generation: Qwen2.5

Pick a size

One row per official Qwen2 size: the VRAM it needs at each precision, the cheapest GPU that holds it today, and the KV cache for a 32K-token context.

ModelParametersNative VRAMFP8 VRAMINT4 VRAMLive GPU fit (native)Est. $/hrKV cache at 32K
Qwen2-0.5B494M1.1 GB0.6 GB0.3 GBRTX 3070$0.088/hr0.38 GB
Qwen2-1.5B-Instruct1.5B3.5 GB1.7 GB0.9 GBRTX 3070$0.088/hr0.88 GB
Qwen2-VL-2B-Instruct2.2B4.9 GB2.5 GB1.2 GBRTX 3070$0.088/hr0.88 GB
Qwen2-7B-Instruct7.6B17.0 GB8.5 GB4.3 GBRTX A5000$0.176/hr1.8 GB
Qwen2-VL-7B-Instruct8.3B18.5 GB9.3 GB4.6 GBRTX A5000$0.176/hr1.8 GB
Qwen2-57B-A14B-Instruct57.4B128 GB64.2 GB32.1 GBAMD MI300X$2.63/hr1.8 GB
Qwen2-72B-Instruct72.7B163 GB81.3 GB40.6 GBAMD MI300X$2.63/hr10.0 GB

VRAM is weights times a flat 1.2 overhead; the KV cache is a separate per-model figure at 16-bit, shown where the architecture is published. See the methodology.

Official models (12)

ModelParametersPublished asVRAM neededLive GPU fitEst. $/hr
Qwen2-0.5B494MBF161.1 GBRTX 3070$0.088/hr
Qwen2-0.5B-Instruct494MBF161.1 GBRTX 3070$0.088/hr
Qwen2-1.5B1.5BBF163.5 GBRTX 3070$0.088/hr
Qwen2-1.5B-Instruct1.5BBF163.5 GBRTX 3070$0.088/hr
Qwen2-VL-2B-Instruct2.2BBF164.9 GBRTX 3070$0.088/hr
Qwen2-7B7.6BBF1617.0 GBRTX A5000$0.176/hr
Qwen2-7B-Instruct7.6BBF1617.0 GBRTX A5000$0.176/hr
Qwen2-VL-7B-Instruct8.3BBF1618.5 GBRTX A5000$0.176/hr
Qwen2-57B-A14B57.4BBF16128 GBAMD MI300X$2.63/hr
Qwen2-57B-A14B-Instruct57.4BBF16128 GBAMD MI300X$2.63/hr
Qwen2-72B72.7BBF16163 GBAMD MI300X$2.63/hr
Qwen2-72B-Instruct72.7BBF16163 GBAMD MI300X$2.63/hr

Quantized versions (1)

ModelParametersPublished asVRAM neededLive GPU fitEst. $/hr
Qwen2-1.5B-Instruct-FP81.5BF8_E4M31.7 GBRTX 5060 Ti$0.180/hr

Fine-tunes and community models (2)

ModelParametersPublished asVRAM neededLive GPU fitEst. $/hr
gte-Qwen2-1.5B-instruct1.8BF327.9 GBRTX 3070$0.088/hr
gte-Qwen2-7B-instruct7.6BF3234.0 GBRTX A6000$0.363/hr

VRAM is for the precision the model is published in: the weights times a flat 1.2 overhead, with the KV cache not included. See the methodology. The fit shown is the lowest-priced single GPU type that holds the model at that precision, or the lowest-priced multi-GPU set (up to 8) when none does.

Run and fine-tune Qwen2

  • How to run Qwen2: engine support, chat template and context flags, with a serving command sized per model.
  • How to fine-tune Qwen2: LoRA and QLoRA memory per model, and the license limits on derivatives.

More from Qwen

Start ComfyUI on a rented GPU.

© 2026 Aquanode. All rights reserved.

All trademarks, logos and brand names are the property of their respective owners.