Can I Run AI locally?

Meta · 8B

Meta's versatile 8B — great quality/speed ratio

Alibaba · 9B

Multimodal Qwen 3.5 mid-size

Microsoft · 14B

Microsoft's reasoning-focused model

OpenAI · 21B

OpenAI's open-weight MoE with configurable reasoning

Mistral AI · 24B

Multimodal Mistral with vision support

Google · 27B

Google's flagship Gemma 3 model

Alibaba · 32B

Best open-source coding model at release

Alibaba · 32B

Qwen 3 flagship dense model

DeepSeek · 32B

R1 reasoning distilled into Qwen 32B — sweet spot

Meta · 70B

Best open model at 70B class

Meta · 109B

MoE with 16 experts, 17B active params

OpenAI · 117B

OpenAI's flagship open-weight MoE — 52.6% SWE-bench

Mistral AI · 123B

Dense 123B coding model — 72.2% SWE-bench Verified

DeepSeek · 671B

Massive MoE reasoning model — 37B active

DeepSeek · 685B

State-of-the-art MoE — 37B active params

Moonshot AI · 1T

1T-param MoE with 384 experts — 32B active, strong agentic coding

Alibaba · 0.8B

Ultra-tiny model for embedded and edge

Meta · 1B

Meta's smallest Llama for edge devices

Google · 1B

Google's tiny Gemma for on-device

Community · 1.1B

Ultralight model for constrained devices

Alibaba · 1.5B

Ultra-lightweight coding model

DeepSeek · 1.5B

Tiny reasoning model distilled from R1

Alibaba · 1.7B

Compact multilingual Qwen 3

Alibaba · 2B

Small multimodal Qwen 3.5

Google · 2B

Google's compact open model

Meta · 3B

Lightweight Llama for mobile and edge

HuggingFace · 3B

Lightweight multilingual reasoning

Microsoft · 3.8B

Microsoft's efficient small model with long context

Microsoft · 3.8B

Lightweight reasoning model

Alibaba · 4B

Compact Qwen 3 for general tasks

Google · 4B

Multimodal Gemma with 128K context

Alibaba · 4B

Small multimodal Qwen 3.5

Mistral AI · 7B

High-quality 7B with sliding window attention

Alibaba · 7B

Strong multilingual and coding capabilities

Alibaba · 7B

Dedicated coding model

DeepSeek · 7B

R1 reasoning distilled into Qwen 7B

Alibaba · 8B

Qwen 3 with thinking mode support

Mistral AI · 8B

Mistral's efficient 8B model

Google · 9B

Google's best mid-size open model

Zhipu AI · 9B

Multilingual model supporting 26 languages with 128K context

NVIDIA · 9B

Hybrid Mamba2 architecture for reasoning

Meta · 11B

Multimodal vision and text model

Google · 12B

Multimodal Gemma with 128K context

Mistral AI · 12B

Multilingual 12B with 128K context

Alibaba · 14B

Excellent quality for its size class

Alibaba · 14B

Strong all-rounder with thinking mode

DeepSeek · 14B

R1 reasoning distilled into Qwen 14B

Liquid AI · 24B

Hybrid MoE with convolution+attention layers — 2.3B active

Mistral AI · 24B

Coding-focused model with 256K context — 68% SWE-bench

Google · 27B

Google's largest Gemma 2 model

Alibaba · 27.8B

Flagship native multimodal Qwen 3.5

Alibaba · 30B

MoE with only 3.3B active — extremely efficient

NVIDIA · 30B

MoE with 1M context and 3B active

Alibaba · 32B

High-quality reasoning and multilingual

LG AI · 32B

Hybrid reasoning, multilingual

Allen AI · 32B

Fully open research model by Allen AI

Cohere · 35B

Optimized for retrieval-augmented generation

Alibaba · 35B

Efficient multimodal MoE with 3B active

Mistral AI · 47B

MoE with 12.9B active params

Alibaba · 72B

Alibaba's flagship open model

Alibaba · 122B

Large multimodal MoE with 10B active

Mistral AI · 141B

Large MoE with 39B active params

Alibaba · 235B

Massive MoE with 22B active — frontier quality

Alibaba · 397B

Largest multimodal Qwen 3.5 MoE

Meta · 400B

Multimodal MoE with 128 experts — 17B active, 1M context

Meta · 405B

Largest open-weight dense model by Meta

Alibaba · 480B

Largest open coding MoE — 35B active

DeepSeek · 671B

Improved V3 with hybrid thinking and tool use