Loading…
Loading…
Models / Qwen3.8
1 model (27B) · ↓ 16.9M downloads a month · from $0.14/h on RTX A4000 right now. Each model's versions do the same job; quantized ones (AWQ, FP8, 4-bit) need less GPU memory and run on cheaper machines.
| Version | Runtime | Weights | GPU memory | Cheapest machine now |
|---|---|---|---|---|
| OriginalQwen/Qwen3.8-27B | vLLM | 56 GB | 66.3 GB | $0.89/h 2x Nvidia A40 |
| FP8Qwen/Qwen3.8-27B-FP8 | vLLM | 31 GB | 38.9 GB | $0.45/h Nvidia A40 |
| NVFP4 (RadixArk)RadixArk/Qwen3.8-27B-NVFP4 | vLLM | 22 GB | 29 GB | $0.34/h 2x RTX 5060 Ti |
| AWQ INT4 (cyankiwi)cyankiwi/Qwen3.8-27B-AWQ-INT4 | vLLM | 21 GB | 28 GB | $0.34/h 2x RTX 5060 Ti |
| NVFP4 (nvidia)nvidia/Qwen3.8-27B-NVFP4 | vLLM | 22 GB | 29 GB | $0.34/h 2x RTX 5060 Ti |
| INT4 (RedHatAI)RedHatAI/Qwen3.8-27B-INT4 | vLLM | 19 GB | 26.2 GB | $0.34/h 2x RTX 5060 Ti |
| MTP NVFP4 (sakamakismile)sakamakismile/Qwen3.8-27B-MTP-NVFP4 | vLLM | 21 GB | 27.5 GB | $0.34/h 2x RTX 5060 Ti |
| NVFP4 (Inferact)Inferact/Qwen3.8-27B-NVFP4 | vLLM | 26 GB | 33.9 GB | $0.45/h Nvidia A40 |
| NVFP4 RTX5090 (gittensor-model-hub)gittensor-model-hub/Qwen3.8-27B-NVFP4-RTX5090 | vLLM | 18 GB | 24.5 GB | $0.34/h 2x RTX 5060 Ti |
| NVFP4 BF16 LMHead (RadixArk)RadixArk/Qwen3.8-27B-NVFP4-BF16-LMHead | vLLM | 24 GB | 31 GB | $0.41/h RTX PRO 4500 |
| Uncensored FP8 (orcarouter)orcarouter/Qwen3.8-27B-Uncensored-FP8Licence to accept | vLLM | 31 GB | ~39.3 GB | $0.45/h Nvidia A40 |
| QUASAR NVFP4 (QUASAR-QAT)QUASAR-QAT/Qwen3.8-27B-QUASAR-NVFP4 | vLLM | 21 GB | 27.5 GB | $0.34/h 2x RTX 5060 Ti |
| Uncensored NVFP4 (orcarouter)orcarouter/Qwen3.8-27B-Uncensored-NVFP4Licence to accept | vLLM | 25 GB | ~31.8 GB | $0.41/h RTX PRO 4500 |
| W4A16 AutoRound GPTQ (Vishva007)Vishva007/Qwen3.8-27B-W4A16-AutoRound-GPTQ | vLLM | 20 GB | 26.9 GB | $0.34/h 2x RTX 5060 Ti |
| FP8 Pessoa (lribeiro)lribeiro/Qwen3.8-27B-FP8-Pessoa | vLLM | 31 GB | 39.3 GB | $0.45/h Nvidia A40 |
| W4A16 AWQ (philbert440)philbert440/Qwen3.8-27B-W4A16-AWQ | vLLM | 20 GB | 26.3 GB | $0.34/h 2x RTX 5060 Ti |
| INT8 W8A16 MTP (lued)lued/Qwen3.8-27B-INT8-W8A16-MTP | vLLM | 32 GB | 39.7 GB | $0.45/h Nvidia A40 |
| AWQ INT4 (avyukth)avyukth/Qwen3.8-27B-AWQ-INT4 | vLLM | 19 GB | 25.2 GB | $0.34/h 2x RTX 5060 Ti |
| FP8 (huginnfork)huginnfork/Qwen3.8-27B-FP8 | vLLM | 38 GB | 47.3 GB | $0.46/h RTX A6000 |
| GPTQ Int4 sym G128 MTP BF16 (SergiioB)SergiioB/Qwen3.8-27B-GPTQ-Int4-sym-G128-MTP-BF16 | vLLM | 20 GB | 26.3 GB | $0.34/h 2x RTX 5060 Ti |
| GPTQ Int4 sym G128 MTP BF16 (mikeinnyc)mikeinnyc/Qwen3.8-27B-GPTQ-Int4-sym-G128-MTP-BF16 | vLLM | 20 GB | 26.3 GB | $0.34/h 2x RTX 5060 Ti |
| Ternary Bonsai 2 27B vllm (fraserprice)fraserprice/Ternary-Bonsai-2-27B-vllm | vLLM | 10 GB | 16 GB | $0.18/h 2x RTX 3060 |
| OrcaSAQ 2 27B (orcarouter)orcarouter/OrcaSAQ-2-27B | vLLM | 12 GB | 18.2 GB | $0.18/h 2x RTX 3060 |
| Escha W2 (EschaLabs)EschaLabs/Qwen3.8-27B-Escha-W2 | vLLM | 10 GB | 15.9 GB | $0.14/h RTX A4000 |
| NVFP4 ONNX (tlwu)tlwu/Qwen3.8-27B-NVFP4-ONNX | ONNX Runtime GenAI | 23 GB | 29.7 GB | $0.18/h Nvidia V100 |
Model details come from Hugging Face and are refreshed daily; prices follow the live offers. GPU memory is GridPup's estimate: the weights, room for an 8,192-token conversation (chat models) and the runtime's own memory. Large models are split across the GPUs of one machine ("4x RTX 4090").