Model registry

Open-source LLMs

The major open-source LLMs anyone can download and run. Each page shows the model's size, its license, how much memory it needs at different precisions, and which GPUs actually fit it.

Kimi K2

1000 B params

Moonshot AI · Modified MIT · released 2025-07-11 · MoE (1T total, 32B active per token)

DeepSeek V3

671 B params

DeepSeek · MIT · released 2024-12-26 · MoE (671B total, 37B active per token)

DeepSeek R1

671 B params

DeepSeek · MIT · released 2025-01-20 · MoE (671B total, 37B active per token)

Llama 3.1 405B

405 B params

Meta · Llama 3.1 Community License · released 2024-07-23 · dense

Mixtral 8x22B

141 B params

Mistral AI · Apache 2.0 · released 2024-04-17 · MoE (141B total, 39B active per token)

Command R+

104 B params

Cohere · CC-BY-NC-4.0 (non-commercial) · released 2024-04-04 · dense

Qwen 2.5 72B

72 B params

Alibaba · Qwen License (72B only) · released 2024-09-19 · dense

Llama 3.1 70B

70 B params

Meta · Llama 3.1 Community License · released 2024-07-23 · dense

Llama 3.3 70B

70 B params

Meta · Llama 3.3 Community License · released 2024-12-06 · dense

Mixtral 8x7B

46.7 B params

Mistral AI · Apache 2.0 · released 2023-12-11 · MoE (46.7B total, 12.9B active per token)

Gemma 2 27B

27 B params

Google · Gemma Terms of Use · released 2024-06-27 · dense

Llama 3.1 8B

8 B params

Meta · Llama 3.1 Community License · released 2024-07-23 · dense

Qwen 2.5 7B

7 B params

Alibaba · Apache 2.0 · released 2024-09-19 · dense