By input price per 1M tokens · prices verified on official pages
| Model | Vendor | Input $/1M | Output $/1M | Context |
|---|---|---|---|---|
| Llama 3 8B Instruct Liteopen | Meta | $0.14 | $0.14 | 8K |
| DeepSeek V4 Flash | DeepSeek | $0.14 | $0.28 | 1M |
| GPT-5.4 Nano | OpenAI | $0.2 | $1.25 | 1.05M |
| Gemini 2.5 Flash | $0.3 | $2.5 | 1M | |
| DeepSeek V4 Proopen | DeepSeek | $0.435 | $0.87 | 1M |
| GPT-5.4 Mini | OpenAI | $0.75 | $4.5 | 1.05M |
| Claude Haiku 4.5 | Anthropic | $1 | $5 | 200K |
| GPT-5.6 Luna | OpenAI | $1 | $6 | 1.05M |
Ranked by monthly $ per GB of VRAM — the spec that decides which models you can run
| Plan | GPU | VRAM | $/GB VRAM | Monthly | |
|---|---|---|---|---|---|
| Basic GPU Dedicated Server - K80DatabaseMart | Nvidia Tesla K80 | 24GB | $5.38 | $129.00/mo | |
| Multi-GPU Dedicated Server - 3xRTX A5000DatabaseMart | 3 x Quadro RTX A5000 | 72GB ×3 | $6.31 | $454.35/mo | |
| Multi-GPU Dedicated Server - 3xRTX A6000DatabaseMart | 3 x Nvidia Quadro RTX A6000 | 144GB ×3 | $8.33 | $1,199.00/mo | |
| GEX44Hetzner | NVIDIA RTX 4000 SFF Ada Generation | 20GB | $10.25 | $205.00/mo | |
| Basic GPU Dedicated Server - T1000DatabaseMart | Nvidia Quadro T1000 | 8GB | $10.41 | $83.30/mo | |
| Professional GPU VPS - A4000DatabaseMart | Quadro RTX A4000 | 16GB | $11.19 | $179.00/mo | |
| Multi-GPU Dedicated Server - 2xRTX A5000DatabaseMart | 2 x Quadro RTX A5000 | 48GB ×2 | $11.23 | $539.00/mo | |
| Enterprise GPU Dedicated Server - RTX A6000DatabaseMart | Nvidia Quadro RTX A6000 | 48GB | $11.44 | $549.00/mo |
APIs win for spiky, low-volume workloads. A flat-rate GPU server wins when you push tokens around the clock — open-weight models like Llama 3.3 70B can run on hardware you rent by the month. Our guide maps model sizes to VRAM requirements and the live cheapest server that clears each bar.