AWS Savings plans can offer discounts of around 30% on the AWS prices shown below. Prices are shown for on-demand instances.
| Model* | vRam | GPU Examples for Purchase | AWS Instance / Cost per hour** | AWS 24X7 monthly *** | AWS
200 hours monthly *** | Features (ratings / 10) | Cloud API Option |
|---|---|---|
Open AI GPT OSS 20B | 16 GB | Data Center GPU: |
L4, 24G - $2600
Consumer GPU:
RTX 3090/4090, 24 GB ($2400) | N. Virginia / Stockholm: G4dn.xlarge ($0.52) Israel: g5.xlarge · 1×A10G (24 GB)
$1.179 | $400 $900 | $100 $240 | Answer Accuracy - 7 Code Generation - 8 Agent - Translation - Vision - No | Bedrock |
| Open AI GPT OSS 120B | 80 GB | Data Center GPU: 2x L40s, 96 GB total ($16,000) Consumer GPU: 4x 4090, 96 GB Total ($9600) | N. Virginia g6.12xlarge ($4.60) Stockholm g5.12xlarge ($6) Israel g5.12xlarge ($6.64) | $3500 $4500 $5000 | $900 $1200 $1300 | Answer Accuracy - 8 Code Generation - 9 Agent - Translation - Vision - No | Bedrock |
| Gemma 3 27 B | 80 GB | Data Center GPU: 2x L40s, 96 GB total ($16,000) (65,000 NIS in Israel) Consumer GPU: Information upon request | N. Virginia g6.12xlarge ($4.60) Stockholm g5.12xlarge ($6) Israel g5.12xlarge ($6.64) | $3500 $4500 $5000 | $900 $1200 $1300 | Answer Accuracy - 7 Code Generation - 8 Agent - Vision - Yes Translation - Yes | OpenRouter |
| Llama 4 Maverick | 640 GB | 8× H100 80 GB ($200,000) | p5.48xlarge ( $55/ hr) | $41,000 | $11,000 | Answer Accuracy - 8 Code Generation - 9 Agent - Yes Vision - Yes Translation - Yes | Bedrock |
| Llama 3.1 8B (Deprecated) | 24 GB | Data Center GPU:
L4, 24G - $2600 Consumer GPU:
RTX 4090, 24 GB ($2400) | N. Virginia g6.xlarge ($0.80) | $600 | $160 | Answer Accuracy - 5 Code Generation - N/A Agent - No Translation - No Vision - No | Bedrock |
* Above models are official releases only
** LLM Instance only. Windows Dashboard and Linux Gateway instances required too.
*** Prices are approximate
Also tested:
Oss 20B: RTX 2000 ADA, RTX A4500 (RTX A4000 didn't pass testing)
Additional Cloud Cost References
For broader cloud cost data (AWS and Runpod GPU comparisons), see source spreadsheet: Business GPT Private AI Costs.xlsx
Note: Tesla L4 & Tesla L40S are different GPUs.