Qwen3-4B-Thinking-2507-FP8: GPU memory and hosting cost
Open weights · 4.41 billion parameters · 262,144-token context · licence apache-2.0
No cloud sells this model per token: to use it you run it yourself, on GPU instances you rent.
Run it yourself
It needs 5.5 GB of GPU memory at fp8, the precision it ships in.
| Cloud | Cheapest instance that fits | GPUs | An hour | A month, around the clock |
|---|---|---|---|---|
| HF Endpoints | aws.nvidia-t4.x1 | 1 × T4 14 GB | $0.50 | $365 |
| AWS | g4dn.xlarge | 1 × T4 16 GB | $0.526 | $384 |
| Azure | Standard_NC4as_T4_v3 | 1 × T4 16 GB | $0.526 | $384 |
| Google Cloud | g2-standard-4 | 1 × L4 24 GB | $0.7068 | $516 |
| Alibaba Cloud | ecs.gn6i-c4g1.xlarge | 1 × T4 16 GB | $1.12 | $816 |
| DeepInfra | a100.1x | 1 × A100 80 GB | $1.49 | $1,088 |
| Crusoe | l40s.1x | 1 × L40S 48 GB | $1.50 | $1,095 |
| Nebius | gpu-l40s-a.1gpu-8vcpu-32gb | 1 × L40S 48 GB | $1.55 | $1,130 |
| OCI | VM.GPU.A10.1 | 1 × A10 24 GB | $2.00 | $1,460 |
| Core42 | h100.1x | 1 × H100 80 GB | $2.50 | $1,825 |
| IBM watsonx | deploy-on-demand.1xL40S | 1 × L40S 48 GB | $4.43 | $3,234 |
| Fireworks AI | h100.1x | 1 × H100 80 GB | $8.00 | $5,840 |
| Together AI | hgx-h100.8gpu | 8 × H100 80 GB | $31.92 | $23,302 |
More from Qwen
- Qwen3.8 27B
- Qwen3 32B
- Qwen3 235B A22B Instruct 2507
- Qwen3.5 397B A17B
- Qwen3 30B A3B Instruct 2507
- Qwen3 235B A22B
Similar size, open weights
- Llama 3.1 8B
- Mistral 7B
- MiniCPM-V 4.5
- NVIDIA-Nemotron-3-Nano-4B-BF16
- Mistral-7B-Instruct-v0.2
- Qwen2.5-VL-7B-Instruct-NVFP4
List prices collected 4 November 2026, refreshed every three hours. No negotiated rates or taxes.