Open source AI

Open Source LLM Picker

Find the right open source model for your project. Filter by task, model size, and license to compare Llama, Mistral, Phi, Gemma, Qwen, and CodeLlama — with VRAM requirements and deployment options.

Task

Size

License

Llama 3.1 8B

8B parameters

License

Meta Community

VRAM

~6 GB

Best For

Lightweight general tasks, chatbots, edge deployment

Deploy Options

OllamavLLMBedrockTogether AI

Llama 3.1 70B

70B parameters

License

Meta Community

VRAM

~40 GB

Best For

Strong reasoning and coding, enterprise workloads

Deploy Options

vLLMBedrockTogether AIAnyscale

Llama 3.1 405B

405B parameters

License

Meta Community

VRAM

~230 GB (multi-GPU)

Best For

Maximum open-source performance, research

Deploy Options

vLLM (multi-node)BedrockTogether AI

Mistral 7B

7B parameters

License

Apache 2.0

VRAM

~5 GB

Best For

Fast inference, EU compliance, resource-constrained environments

Deploy Options

OllamavLLMBedrockHugging Face

Mixtral 8x7B

46.7B (MoE) parameters

License

Apache 2.0

VRAM

~26 GB

Best For

Multilingual tasks, efficient MoE architecture, coding

Deploy Options

vLLMBedrockTogether AIFireworks

Phi-3 Mini

3.8B parameters

License

MIT

VRAM

~3 GB

Best For

Edge and mobile deployment, on-device AI, lightweight tasks

Deploy Options

OllamaONNXHugging FaceAzure

Gemma 2

9B / 27B parameters

License

Gemma License

VRAM

~6-18 GB

Best For

Research, instruction following, Google ecosystem integration

Deploy Options

OllamavLLMVertex AIHugging Face

Qwen 2

7B / 72B parameters

License

Apache 2.0

VRAM

~5-42 GB

Best For

Multilingual (strong CJK), coding, math reasoning

Deploy Options

OllamavLLMHugging FaceTogether AI

CodeLlama

7B / 13B / 34B parameters

License

Meta Community

VRAM

~5-20 GB

Best For

Code generation, code completion, programming-specific tasks

Deploy Options

OllamavLLMBedrockTogether AI

Deploy LLMs on real AWS

Deploy open source models on Amazon Bedrock and SageMaker in guided, hands-on missions. Real infrastructure, real learning.

Start building free →