Amazon SageMaker
Deploy open or fine-tuned models to managed AWS endpoints, billed per instance-hour.
Overview
Amazon SageMaker AI lets you deploy open-weight models, or your own fine-tuned versions, to managed endpoints that run on instances reserved for you. Popular open models deploy in a few clicks from SageMaker JumpStart. You pay per instance-hour while the endpoint runs, and Savings Plans cut the cost for committed use.
Specifications
| Hosting type | Dedicated endpoint |
|---|---|
| Model families | Llama, Mistral, DeepSeek, Qwen, Gemma, gpt-oss |
| LLM models | Llama 3.1 8B Instruct, Llama 3.3 70B Instruct, Mistral NeMo, Mistral Small 3.1, DeepSeek-R1, DeepSeek-V3.1, Qwen2.5 72B Instruct, Qwen2.5-Coder 32B Instruct, Qwen3 32B, Qwen3 30B-A3B, Gemma 2 9B, Gemma 2 27B, Gemma 3 4B, Gemma 3 27B, gpt-oss-120b, gpt-oss-20b |
| Pricing range | From US$0.736 |
| Pricing model | Per GPU-hour, Reserved or committed capacity |
| Free tier available | No |
| Data center locations | United States, Canada, Mexico, Brazil, Ireland, United Kingdom, Germany, France, Sweden, Italy, Spain, Switzerland, Israel, Bahrain, United Arab Emirates, South Africa, India, Japan, South Korea, Singapore, Australia, Indonesia, Malaysia, Thailand, Taiwan, Hong Kong SAR China, New Zealand, China |
| Hardware / GPU options | NVIDIA T4, A10G, L4, L40S, A100, H100, H200 and B200 GPUs, plus AWS Inferentia2 accelerators. |
| VRAM & memory | 16 GB – 180 GB |
| Maximum model size | 671B |
| OpenAI-compatible API | No |
News about Amazon AWS
All news →LLM Hosting Roundup: Anthropic’s capacity crunch, Vera Rubin and DeepSeek’s $15B (23 Oct 2026)
Anthropic's hunt for cloud capacity, NVIDIA's DSX POD and new Vera Rubin inference benchmarks, Dell's first Groq 3 rack, DeepSeek's reported $15B raise and Reflection's open-weights Beam model.
