Meta
Llama 3.1 8B Instruct
NIMgated
Strong general assistant that fits on a single 24 GB GPU.
- Parameters
- 8.03B
- Max context
- 128K
- Recommended
- g6e.xlarge
Model catalog
Open-weight models, ready to deploy.
Each model is sized from its published architecture. The recommended instance is the cheapest NVIDIA GPU instance that holds the model at 16K context with 8 concurrent requests in BF16.
Meta
Llama 3.3 70B Instruct
NIMgated
Frontier-class open model for multi-GPU instances.
- Parameters
- 70.6B
- Max context
- 128K
- Recommended
- p4d.24xlarge
NVIDIA
Llama 3.1 Nemotron Nano 8B
Reasoning-tuned 8B model with a switchable thinking mode.
- Parameters
- 8.03B
- Max context
- 128K
- Recommended
- g6e.xlarge
Qwen
Qwen2.5 7B Instruct
Efficient multilingual model with a very small KV cache.
- Parameters
- 7.62B
- Max context
- 32K
- Recommended
- g6e.xlarge
Qwen
Qwen2.5 Coder 7B Instruct
Code completion, review and generation.
- Parameters
- 7.62B
- Max context
- 32K
- Recommended
- g6e.xlarge
Qwen
Qwen2.5 14B Instruct
Mid-size model; a good fit for a 48 GB GPU.
- Parameters
- 14.7B
- Max context
- 32K
- Recommended
- g6.12xlarge
Qwen
Qwen2.5 32B Instruct
High quality at a fraction of 70B serving cost.
- Parameters
- 32.8B
- Max context
- 32K
- Recommended
- g6e.12xlarge
Mistral AI
Mistral 7B Instruct v0.3
gated
Fast, permissively licensed 7B assistant.
- Parameters
- 7.25B
- Max context
- 32K
- Recommended
- g6e.xlarge
Mistral AI
Mixtral 8x7B Instruct
gated
Mixture-of-experts: 47B weights, ~13B active per token.
- Parameters
- 46.7B (12.9B act.)
- Max context
- 32K
- Recommended
- g6e.12xlarge
Gemma 2 9B Instruct
gated
Compact model with strong writing quality.
- Parameters
- 9.24B
- Max context
- 8K
- Recommended
- g4dn.12xlarge
Microsoft
Phi-3.5 mini Instruct
Small model that runs well on a 16 GB T4.
- Parameters
- 3.82B
- Max context
- 128K
- Recommended
- g6.12xlarge
DeepSeek
DeepSeek R1 Distill Qwen 7B
Chain-of-thought reasoning distilled into a 7B model.
- Parameters
- 7.62B
- Max context
- 128K
- Recommended
- g6e.xlarge
DeepSeek
DeepSeek R1 Distill Llama 70B
Large reasoning model for 4–8 GPU instances.
- Parameters
- 70.6B
- Max context
- 128K
- Recommended
- p4d.24xlarge
Gated models require you to accept the publisher's license on Hugging Face and add your own access token. AWSBee does not redistribute model weights; your instance downloads them directly from the publisher's repository.