Local LLMs

Run AI on your own machine.

111 open-weight models, from SmolLM2 135M at 271 MB to DeepSeek-V3 671B at 404 GB. Pick one that fits your memory, then run it with a single command. Private, offline and free.

Install one, then pick a model below and copy its command. A model needs roughly its download size plus 20% in free RAM or GPU memory.

Fits in

111 models · memory figures are estimates for the default 4-bit download.

SmolLM2 135M

Hugging Face · Apache 2.0

Parameters: 135M
Download: 271 MB
Context: 8K

Dense modelNeeds about 1 GB RAM

Tiny models that run almost anywhere.

Tools

$ollama run smollm2:135m

Gemma 3 270M

Google · Gemma Terms of Use

Parameters: 270M
Download: 292 MB
Context: 32K

Dense modelNeeds about 1 GB RAM

Lightweight multimodal models that run on a single GPU.

$ollama run gemma3:270m

Granite 4 350M

IBM · Apache 2.0

Parameters: 350M
Download: 708 MB
Context: 32K

Dense modelNeeds about 1 GB RAM

Efficient hybrid models built for enterprise and edge.

Tools

$ollama run granite4:350m

SmolLM2 360M

Hugging Face · Apache 2.0

Parameters: 360M
Download: 726 MB
Context: 8K

Dense modelNeeds about 1 GB RAM

Tiny models that run almost anywhere.

Tools

$ollama run smollm2:360m

Qwen2.5 Coder 0.5B

Alibaba · Apache 2.0

Parameters: 0.5B
Download: 398 MB
Context: 32K

Dense modelNeeds about 1 GB RAM

Code generation, reasoning and fixing in 40+ languages.

ToolsCode

$ollama run qwen2.5-coder:0.5b

Qwen 3 0.6B

Alibaba · Apache 2.0

Parameters: 0.6B
Download: 523 MB
Context: 40K

Dense modelNeeds about 1 GB RAM

Dense and MoE models with switchable thinking.

ToolsThinking

$ollama run qwen3:0.6b

Qwen 3.5 0.8B

Alibaba · Apache 2.0

Parameters: 0.8B
Download: 1.2 GB
Context: 256K

Dense modelNeeds about 2 GB RAM

A full family from tiny to huge, multimodal with thinking.

ToolsThinkingVision

$ollama run qwen3.5:0.8b

Gemma 3 1B

Google · Gemma Terms of Use

Parameters: 1B
Download: 806 MB
Context: 32K

Dense modelNeeds about 1 GB RAM

Lightweight multimodal models that run on a single GPU.

$ollama run gemma3:1b

Llama 3.2 1B

Meta · Llama 3.2 Community License

Parameters: 1B
Download: 1.3 GB
Context: 128K

Dense modelNeeds about 2 GB RAM

Small models for on-device chat, summaries and tools.

Tools

$ollama run llama3.2:1b

Falcon 3 1B

TII · Falcon 3 TII License

Parameters: 1B
Download: 1.8 GB
Context: 8K

Dense modelNeeds about 3 GB RAM

Small, efficient models from the Technology Innovation Institute.

$ollama run falcon3:1b

Granite 4 1B

IBM · Apache 2.0

Parameters: 1B
Download: 3.3 GB
Context: 128K

Dense modelNeeds about 4 GB RAM

Efficient hybrid models built for enterprise and edge.

Tools

$ollama run granite4:1b

TinyLlama 1.1B

TinyLlama project · Apache 2.0

Parameters: 1.1B
Download: 638 MB
Context: 2K

Dense modelNeeds about 1 GB RAM

A compact 1.1B model trained on 3 trillion tokens.

$ollama run tinyllama:1.1b

LFM2.5 Thinking 1.2B

Liquid AI · LFM Open License

Parameters: 1.2B
Download: 731 MB
Context: 125K

Dense modelNeeds about 1 GB RAM

A tiny reasoning model built for on-device use.

ToolsThinking

$ollama run lfm2.5-thinking:1.2b

Qwen2.5 Coder 1.5B

Alibaba · Apache 2.0

Parameters: 1.5B
Download: 986 MB
Context: 32K

Dense modelNeeds about 2 GB RAM

Code generation, reasoning and fixing in 40+ languages.

ToolsCode

$ollama run qwen2.5-coder:1.5b

DeepSeek-R1 1.5B

DeepSeek · MIT

Parameters: 1.5B
Download: 1.1 GB
Context: 128K

Dense modelNeeds about 2 GB RAM

Open reasoning models, from small distills to the full 671B.

ToolsThinking

$ollama run deepseek-r1:1.5b

Qwen 3 1.7B

Alibaba · Apache 2.0

Parameters: 1.7B
Download: 1.4 GB
Context: 40K

Dense modelNeeds about 2 GB RAM

Dense and MoE models with switchable thinking.

ToolsThinking

$ollama run qwen3:1.7b

SmolLM2 1.7B

Hugging Face · Apache 2.0

Parameters: 1.7B
Download: 1.8 GB
Context: 8K

Dense modelNeeds about 3 GB RAM

Tiny models that run almost anywhere.

Tools

$ollama run smollm2:1.7b

Moondream 1.8B

Moondream · Apache 2.0

Parameters: 1.8B
Download: 1.7 GB
Context: 2K

Dense modelNeeds about 3 GB RAM

A tiny vision model for edge devices.

Vision

$ollama run moondream:1.8b

Qwen3 VL 2B

Alibaba · Apache 2.0

Parameters: 2B
Download: 1.9 GB
Context: 256K

Dense modelNeeds about 3 GB RAM

Vision-language models for images, documents and video.

ToolsThinkingVision

$ollama run qwen3-vl:2b

Qwen 3.5 2B

Alibaba · Apache 2.0

Parameters: 2B
Download: 2.7 GB
Context: 256K

Dense modelNeeds about 4 GB RAM

A full family from tiny to huge, multimodal with thinking.

ToolsThinkingVision

$ollama run qwen3.5:2b

Gemma 4 E2B

Google · Apache 2.0

Parameters: E2B
Download: 4.6 GB
Context: 128K

Dense modelNeeds about 6 GB RAM

Google's newest open models with vision, audio, tools and thinking.

ToolsThinkingVision

$ollama run gemma4:e2b

Gemma 3n E2B

Google · Gemma Terms of Use

Parameters: E2B
Download: 5.6 GB
Context: 32K

Dense modelNeeds about 7 GB RAM

Built for phones and laptops: big-model quality in a small memory footprint.

$ollama run gemma3n:e2b

StarCoder2 3B

BigCode · BigCode OpenRAIL-M

Parameters: 3B
Download: 1.7 GB
Context: 16K

Dense modelNeeds about 3 GB RAM

Transparent code models trained on The Stack v2.

Code

$ollama run starcoder2:3b

Qwen2.5 Coder 3B

Alibaba · Apache 2.0

Parameters: 3B
Download: 1.9 GB
Context: 32K

Dense modelNeeds about 3 GB RAM

Code generation, reasoning and fixing in 40+ languages.

ToolsCode

$ollama run qwen2.5-coder:3b

Llama 3.2 3B

Meta · Llama 3.2 Community License

Parameters: 3B
Download: 2.0 GB
Context: 128K

Dense modelNeeds about 3 GB RAM

Small models for on-device chat, summaries and tools.

Tools

$ollama run llama3.2:3b

Hermes 3 3B

Nous Research · Llama 3 Community License

Parameters: 3B
Download: 2.0 GB
Context: 128K

Dense modelNeeds about 3 GB RAM

Steerable, role-play friendly fine-tunes of Llama.

Tools

$ollama run hermes3:3b

Falcon 3 3B

TII · Falcon 3 TII License

Parameters: 3B
Download: 2.0 GB
Context: 32K

Dense modelNeeds about 3 GB RAM

Small, efficient models from the Technology Innovation Institute.

$ollama run falcon3:3b

Granite 4 3B

IBM · Apache 2.0

Parameters: 3B
Download: 2.1 GB
Context: 128K

Dense modelNeeds about 3 GB RAM

Efficient hybrid models built for enterprise and edge.

Tools

$ollama run granite4:3b

Cogito 3B

Deep Cogito · Llama and Qwen licenses

Parameters: 3B
Download: 2.2 GB
Context: 128K

Dense modelNeeds about 3 GB RAM

Hybrid reasoning models that can answer directly or think first.

Tools

$ollama run cogito:3b

Ministral 3 3B

Mistral AI · Apache 2.0

Parameters: 3B
Download: 3.0 GB
Context: 256K

Dense modelNeeds about 4 GB RAM

Edge models with vision and tools.

ToolsVision

$ollama run ministral-3:3b

Phi-4 mini 3.8B

Microsoft · MIT

Parameters: 3.8B
Download: 2.5 GB
Context: 128K

Dense modelNeeds about 3 GB RAM

Compact, multilingual, with function calling.

Tools

$ollama run phi4-mini:3.8b

Qwen 3 4B

Alibaba · Apache 2.0

Parameters: 4B
Download: 2.5 GB
Context: 256K

Dense modelNeeds about 3 GB RAM

Dense and MoE models with switchable thinking.

ToolsThinking

$ollama run qwen3:4b

Qwen 3.5 4B

Alibaba · Apache 2.0

Parameters: 4B
Download: 3.3 GB
Context: 256K

Dense modelNeeds about 4 GB RAM

A full family from tiny to huge, multimodal with thinking.

ToolsThinkingVision

$ollama run qwen3.5:4b

Qwen3 VL 4B

Alibaba · Apache 2.0

Parameters: 4B
Download: 3.3 GB
Context: 256K

Dense modelNeeds about 4 GB RAM

Vision-language models for images, documents and video.

ToolsThinkingVision

$ollama run qwen3-vl:4b

Gemma 3 4B

Google · Gemma Terms of Use

Parameters: 4B
Download: 3.4 GB
Context: 128K

Dense modelNeeds about 5 GB RAM

Lightweight multimodal models that run on a single GPU.

Vision

$ollama run gemma3:4b

Gemma 4 E4B

Google · Apache 2.0

Parameters: E4B
Download: 6.6 GB
Context: 128K

Dense modelNeeds about 8 GB RAM

Google's newest open models with vision, audio, tools and thinking.

ToolsThinkingVision

$ollama run gemma4:e4b

Gemma 3n E4B

Google · Gemma Terms of Use

Parameters: E4B
Download: 7.5 GB
Context: 32K

Dense modelNeeds about 9 GB RAM

Built for phones and laptops: big-model quality in a small memory footprint.

$ollama run gemma3n:e4b

StarCoder2 7B

BigCode · BigCode OpenRAIL-M

Parameters: 7B
Download: 4.0 GB
Context: 16K

Dense modelNeeds about 5 GB RAM

Transparent code models trained on The Stack v2.

Code

$ollama run starcoder2:7b

Granite 4 7B

IBM · Apache 2.0

Parameters: 7B
Download: 4.2 GB
Context: 1M

1B active per tokenNeeds about 6 GB RAM

Efficient hybrid models built for enterprise and edge.

Tools

$ollama run granite4:7b-a1b-h

Mistral 7B

Mistral AI · Apache 2.0

Parameters: 7B
Download: 4.4 GB
Context: 32K

Dense modelNeeds about 6 GB RAM

The classic, efficient 7B model.

Tools

$ollama run mistral:7b

OLMo 2 7B

Ai2 · Apache 2.0

Parameters: 7B
Download: 4.5 GB
Context: 4K

Dense modelNeeds about 6 GB RAM

Fully open models: weights, data and training code.

$ollama run olmo2:7b

Falcon 3 7B

TII · Falcon 3 TII License

Parameters: 7B
Download: 4.6 GB
Context: 32K

Dense modelNeeds about 6 GB RAM

Small, efficient models from the Technology Innovation Institute.

$ollama run falcon3:7b

Qwen2.5 Coder 7B

Alibaba · Apache 2.0

Parameters: 7B
Download: 4.7 GB
Context: 32K

Dense modelNeeds about 6 GB RAM

Code generation, reasoning and fixing in 40+ languages.

ToolsCode

$ollama run qwen2.5-coder:7b

DeepSeek-R1 7B

DeepSeek · MIT

Parameters: 7B
Download: 4.7 GB
Context: 128K

Dense modelNeeds about 6 GB RAM

Open reasoning models, from small distills to the full 671B.

ToolsThinking

$ollama run deepseek-r1:7b

Hermes 3 8B

Nous Research · Llama 3 Community License

Parameters: 8B
Download: 4.7 GB
Context: 128K

Dense modelNeeds about 6 GB RAM

Steerable, role-play friendly fine-tunes of Llama.

Tools

$ollama run hermes3:8b

Llama 3.1 8B

Meta · Llama 3.1 Community License

Parameters: 8B
Download: 4.9 GB
Context: 128K

Dense modelNeeds about 6 GB RAM

Meta's general models with tool calling and 128K context.

Tools

$ollama run llama3.1:8b

Cogito 8B

Deep Cogito · Llama and Qwen licenses

Parameters: 8B
Download: 4.9 GB
Context: 128K

Dense modelNeeds about 6 GB RAM

Hybrid reasoning models that can answer directly or think first.

Tools

$ollama run cogito:8b

Qwen 3 8B

Alibaba · Apache 2.0

Parameters: 8B
Download: 5.2 GB
Context: 40K

Dense modelNeeds about 7 GB RAM

Dense and MoE models with switchable thinking.

ToolsThinking

$ollama run qwen3:8b

DeepSeek-R1 8B

DeepSeek · MIT

Parameters: 8B
Download: 5.2 GB
Context: 128K

Dense modelNeeds about 7 GB RAM

Open reasoning models, from small distills to the full 671B.

ToolsThinking

$ollama run deepseek-r1:8b

MiniCPM-V 8B

OpenBMB · MiniCPM Model License

Parameters: 8B
Download: 5.5 GB
Context: 32K

Dense modelNeeds about 7 GB RAM

Efficient vision-language model for images and documents.

Vision

$ollama run minicpm-v:8b

Ministral 3 8B

Mistral AI · Apache 2.0

Parameters: 8B
Download: 6.0 GB
Context: 256K

Dense modelNeeds about 8 GB RAM

Edge models with vision and tools.

ToolsVision

$ollama run ministral-3:8b

Qwen3 VL 8B

Alibaba · Apache 2.0

Parameters: 8B
Download: 6.1 GB
Context: 256K

Dense modelNeeds about 8 GB RAM

Vision-language models for images, documents and video.

ToolsThinkingVision

$ollama run qwen3-vl:8b

Qwen 3.5 9B

Alibaba · Apache 2.0

Parameters: 9B
Download: 6.6 GB
Context: 256K

Dense modelNeeds about 8 GB RAM

A full family from tiny to huge, multimodal with thinking.

ToolsThinkingVision

$ollama run qwen3.5:9b

Falcon 3 10B

TII · Falcon 3 TII License

Parameters: 10B
Download: 6.3 GB
Context: 32K

Dense modelNeeds about 8 GB RAM

Small, efficient models from the Technology Innovation Institute.

$ollama run falcon3:10b

Llama 3.2 Vision 11B

Meta · Llama 3.2 Community License

Parameters: 11B
Download: 7.8 GB
Context: 128K

Dense modelNeeds about 10 GB RAM

Image reasoning, captioning and visual Q&A.

Vision

$ollama run llama3.2-vision:11b

Mistral NeMo 12B

Mistral AI · Apache 2.0

Parameters: 12B
Download: 7.1 GB
Context: 1000K

Dense modelNeeds about 9 GB RAM

A 12B model with 128K context, built with NVIDIA.

Tools

$ollama run mistral-nemo:12b

Gemma 4 12B

Google · Apache 2.0

Parameters: 12B
Download: 7.7 GB
Context: 256K

Dense modelNeeds about 10 GB RAM

Google's newest open models with vision, audio, tools and thinking.

ToolsThinkingVision

$ollama run gemma4:12b

Gemma 3 12B

Google · Gemma Terms of Use

Parameters: 12B
Download: 8.2 GB
Context: 128K

Dense modelNeeds about 10 GB RAM

Lightweight multimodal models that run on a single GPU.

Vision

$ollama run gemma3:12b

OLMo 2 13B

Ai2 · Apache 2.0

Parameters: 13B
Download: 8.4 GB
Context: 4K

Dense modelNeeds about 11 GB RAM

Fully open models: weights, data and training code.

$ollama run olmo2:13b

Qwen2.5 Coder 14B

Alibaba · Apache 2.0

Parameters: 14B
Download: 9.0 GB
Context: 32K

Dense modelNeeds about 11 GB RAM

Code generation, reasoning and fixing in 40+ languages.

ToolsCode

$ollama run qwen2.5-coder:14b

DeepSeek-R1 14B

DeepSeek · MIT

Parameters: 14B
Download: 9.0 GB
Context: 128K

Dense modelNeeds about 11 GB RAM

Open reasoning models, from small distills to the full 671B.

ToolsThinking

$ollama run deepseek-r1:14b

Cogito 14B

Deep Cogito · Llama and Qwen licenses

Parameters: 14B
Download: 9.0 GB
Context: 128K

Dense modelNeeds about 11 GB RAM

Hybrid reasoning models that can answer directly or think first.

Tools

$ollama run cogito:14b

Phi-4 14B

Microsoft · MIT

Parameters: 14B
Download: 9.1 GB
Context: 16K

Dense modelNeeds about 11 GB RAM

A small model that is strong at reasoning and maths.

$ollama run phi4:14b

Ministral 3 14B

Mistral AI · Apache 2.0

Parameters: 14B
Download: 9.1 GB
Context: 256K

Dense modelNeeds about 11 GB RAM

Edge models with vision and tools.

ToolsVision

$ollama run ministral-3:14b

Qwen 3 14B

Alibaba · Apache 2.0

Parameters: 14B
Download: 9.3 GB
Context: 40K

Dense modelNeeds about 12 GB RAM

Dense and MoE models with switchable thinking.

ToolsThinking

$ollama run qwen3:14b

Phi-4 reasoning 14B

Microsoft · MIT

Parameters: 14B
Download: 11 GB
Context: 32K

Dense modelNeeds about 14 GB RAM

Phi-4 tuned for step-by-step reasoning.

$ollama run phi4-reasoning:14b

StarCoder2 15B

BigCode · BigCode OpenRAIL-M

Parameters: 15B
Download: 9.1 GB
Context: 16K

Dense modelNeeds about 11 GB RAM

Transparent code models trained on The Stack v2.

Code

$ollama run starcoder2:15b

gpt-oss 20B

OpenAI · Apache 2.0

Parameters: 20B
Download: 14 GB
Context: 128K

3.6B active per tokenNeeds about 17 GB RAM

OpenAI's open-weight reasoning models for agents and tools.

ToolsThinking

$ollama run gpt-oss:20b

Codestral 22B

Mistral AI · Mistral Non-Production License

Parameters: 22B
Download: 13 GB
Context: 32K

Dense modelNeeds about 16 GB RAM

Code completion and generation in 80+ languages.

Code

$ollama run codestral:22b

Magistral 24B

Mistral AI · Apache 2.0

Parameters: 24B
Download: 14 GB
Context: 39K

Dense modelNeeds about 17 GB RAM

Mistral's transparent reasoning model.

ToolsThinking

$ollama run magistral:24b

Mistral Small 3.2 24B

Mistral AI · Apache 2.0

Parameters: 24B
Download: 15 GB
Context: 128K

Dense modelNeeds about 18 GB RAM

Fast 24B model with vision and reliable tool use.

ToolsVision

$ollama run mistral-small3.2:24b

Devstral Small 2 24B

Mistral AI · See model page

Parameters: 24B
Download: 15 GB
Context: 384K

Dense modelNeeds about 18 GB RAM

An agentic model for software engineering tasks.

ToolsCodeVision

$ollama run devstral-small-2:24b

Gemma 4 26B

Google · Apache 2.0

Parameters: 26B
Download: 16 GB
Context: 256K

4B active per tokenNeeds about 20 GB RAM

Google's newest open models with vision, audio, tools and thinking.

ToolsThinkingVision

$ollama run gemma4:26b

Gemma 3 27B

Google · Gemma Terms of Use

Parameters: 27B
Download: 17 GB
Context: 128K

Dense modelNeeds about 21 GB RAM

Lightweight multimodal models that run on a single GPU.

Vision

$ollama run gemma3:27b

Qwen 3.5 27B

Alibaba · Apache 2.0

Parameters: 27B
Download: 17 GB
Context: 256K

Dense modelNeeds about 21 GB RAM

A full family from tiny to huge, multimodal with thinking.

ToolsThinkingVision

$ollama run qwen3.5:27b

Qwen 3.8 27B

Alibaba · Apache 2.0

Parameters: 27B
Download: 18 GB
Context: 256K

Dense modelNeeds about 22 GB RAM

The latest dense Qwen with vision, tools and thinking.

ToolsThinkingVision

$ollama run qwen3.8:27b

Qwen 3.6 27B

Alibaba · Apache 2.0

Parameters: 27B
Download: 18 GB
Context: 256K

Dense modelNeeds about 22 GB RAM

Qwen with strong coding and agent skills, dense and MoE.

ToolsThinkingVision

$ollama run qwen3.6:27b

Qwen 3 30B

Alibaba · Apache 2.0

Parameters: 30B
Download: 19 GB
Context: 256K

3B active per tokenNeeds about 23 GB RAM

Dense and MoE models with switchable thinking.

ToolsThinking

$ollama run qwen3:30b

Qwen3 Coder 30B

Alibaba · Apache 2.0

Parameters: 30B
Download: 19 GB
Context: 256K

3B active per tokenNeeds about 23 GB RAM

Agentic coding models for long, multi-step tasks.

ToolsCode

$ollama run qwen3-coder:30b

GLM-4.7-Flash 30B

Z.ai · MIT

Parameters: 30B
Download: 19 GB
Context: 198K

3B active per tokenNeeds about 23 GB RAM

A fast, efficient model for coding and agents.

ToolsThinking

$ollama run glm-4.7-flash:latest

Qwen3 VL 30B

Alibaba · Apache 2.0

Parameters: 30B
Download: 20 GB
Context: 256K

3B active per tokenNeeds about 24 GB RAM

Vision-language models for images, documents and video.

ToolsThinkingVision

$ollama run qwen3-vl:30b

Gemma 4 31B

Google · Apache 2.0

Parameters: 31B
Download: 19 GB
Context: 256K

Dense modelNeeds about 23 GB RAM

Google's newest open models with vision, audio, tools and thinking.

ToolsThinkingVision

$ollama run gemma4:31b

Granite 4 32B

IBM · Apache 2.0

Parameters: 32B
Download: 19 GB
Context: 1M

9B active per tokenNeeds about 23 GB RAM

Efficient hybrid models built for enterprise and edge.

Tools

$ollama run granite4:32b-a9b-h

Qwen 3 32B

Alibaba · Apache 2.0

Parameters: 32B
Download: 20 GB
Context: 40K

Dense modelNeeds about 24 GB RAM

Dense and MoE models with switchable thinking.

ToolsThinking

$ollama run qwen3:32b

Qwen2.5 Coder 32B

Alibaba · Apache 2.0

Parameters: 32B
Download: 20 GB
Context: 32K

Dense modelNeeds about 24 GB RAM

Code generation, reasoning and fixing in 40+ languages.

ToolsCode

$ollama run qwen2.5-coder:32b

QwQ 32B

Alibaba · Apache 2.0

Parameters: 32B
Download: 20 GB
Context: 40K

Dense modelNeeds about 24 GB RAM

A medium-sized reasoning model that thinks before answering.

Tools

$ollama run qwq:32b

DeepSeek-R1 32B

DeepSeek · MIT

Parameters: 32B
Download: 20 GB
Context: 128K

Dense modelNeeds about 24 GB RAM

Open reasoning models, from small distills to the full 671B.

ToolsThinking

$ollama run deepseek-r1:32b

Cogito 32B

Deep Cogito · Llama and Qwen licenses

Parameters: 32B
Download: 20 GB
Context: 128K

Dense modelNeeds about 24 GB RAM

Hybrid reasoning models that can answer directly or think first.

Tools

$ollama run cogito:32b

Qwen3 VL 32B

Alibaba · Apache 2.0

Parameters: 32B
Download: 21 GB
Context: 256K

Dense modelNeeds about 26 GB RAM

Vision-language models for images, documents and video.

ToolsThinkingVision

$ollama run qwen3-vl:32b

Qwen 3.5 35B

Alibaba · Apache 2.0

Parameters: 35B
Download: 22 GB
Context: 256K

3B active per tokenNeeds about 27 GB RAM

A full family from tiny to huge, multimodal with thinking.

ToolsThinkingVision

$ollama run qwen3.5:35b

Qwen 3.6 35B

Alibaba · Apache 2.0

Parameters: 35B
Download: 23 GB
Context: 256K

3B active per tokenNeeds about 28 GB RAM

Qwen with strong coding and agent skills, dense and MoE.

ToolsThinkingVision

$ollama run qwen3.6:35b

Mixtral 47B

Mistral AI · Apache 2.0

Parameters: 47B
Download: 26 GB
Context: 32K

13B active per tokenNeeds about 32 GB RAM

Sparse mixture-of-experts models.

Tools

$ollama run mixtral:8x7b

Hermes 3 70B

Nous Research · Llama 3 Community License

Parameters: 70B
Download: 40 GB
Context: 128K

Dense modelNeeds about 48 GB RAM

Steerable, role-play friendly fine-tunes of Llama.

Tools

$ollama run hermes3:70b

Llama 3.3 70B

Meta · Llama 3.3 Community License

Parameters: 70B
Download: 43 GB
Context: 128K

Dense modelNeeds about 52 GB RAM

70B quality close to Llama 3.1 405B.

Tools

$ollama run llama3.3:70b

Llama 3.1 70B

Meta · Llama 3.1 Community License

Parameters: 70B
Download: 43 GB
Context: 128K

Dense modelNeeds about 52 GB RAM

Meta's general models with tool calling and 128K context.

Tools

$ollama run llama3.1:70b

DeepSeek-R1 70B

DeepSeek · MIT

Parameters: 70B
Download: 43 GB
Context: 128K

Dense modelNeeds about 52 GB RAM

Open reasoning models, from small distills to the full 671B.

ToolsThinking

$ollama run deepseek-r1:70b

Cogito 70B

Deep Cogito · Llama and Qwen licenses

Parameters: 70B
Download: 43 GB
Context: 128K

Dense modelNeeds about 52 GB RAM

Hybrid reasoning models that can answer directly or think first.

Tools

$ollama run cogito:70b

Llama 3.2 Vision 90B

Meta · Llama 3.2 Community License

Parameters: 90B
Download: 55 GB
Context: 128K

Dense modelNeeds about 66 GB RAM

Image reasoning, captioning and visual Q&A.

Vision

$ollama run llama3.2-vision:90b

Llama 4 109B

Meta · Llama 4 Community License

Parameters: 109B
Download: 67 GB
Context: 10M

17B active per tokenNeeds about 81 GB RAM

Natively multimodal mixture-of-experts models.

ToolsVision

$ollama run llama4:scout

gpt-oss 120B

OpenAI · Apache 2.0

Parameters: 120B
Download: 65 GB
Context: 128K

5.1B active per tokenNeeds about 78 GB RAM

OpenAI's open-weight reasoning models for agents and tools.

ToolsThinking

$ollama run gpt-oss:120b

Nemotron 3 Super 120B

NVIDIA · NVIDIA Evaluation License

Parameters: 120B
Download: 87 GB
Context: 256K

12B active per tokenNeeds about 105 GB RAM

A large hybrid reasoning model for agents.

ToolsThinking

$ollama run nemotron-3-super:120b

Qwen 3.5 122B

Alibaba · Apache 2.0

Parameters: 122B
Download: 81 GB
Context: 256K

10B active per tokenNeeds about 98 GB RAM

A full family from tiny to huge, multimodal with thinking.

ToolsThinkingVision

$ollama run qwen3.5:122b

Mixtral 141B

Mistral AI · Apache 2.0

Parameters: 141B
Download: 80 GB
Context: 64K

39B active per tokenNeeds about 96 GB RAM

Sparse mixture-of-experts models.

Tools

$ollama run mixtral:8x22b

Qwen 3 235B

Alibaba · Apache 2.0

Parameters: 235B
Download: 142 GB
Context: 256K

22B active per tokenNeeds about 171 GB RAM

Dense and MoE models with switchable thinking.

ToolsThinking

$ollama run qwen3:235b

Qwen3 VL 235B

Alibaba · Apache 2.0

Parameters: 235B
Download: 143 GB
Context: 256K

22B active per tokenNeeds about 172 GB RAM

Vision-language models for images, documents and video.

ToolsThinkingVision

$ollama run qwen3-vl:235b

Llama 4 400B

Meta · Llama 4 Community License

Parameters: 400B
Download: 245 GB
Context: 1M

17B active per tokenNeeds about 294 GB RAM

Natively multimodal mixture-of-experts models.

ToolsVision

$ollama run llama4:maverick

Hermes 3 405B

Nous Research · Llama 3 Community License

Parameters: 405B
Download: 229 GB
Context: 128K

Dense modelNeeds about 275 GB RAM

Steerable, role-play friendly fine-tunes of Llama.

Tools

$ollama run hermes3:405b

Llama 3.1 405B

Meta · Llama 3.1 Community License

Parameters: 405B
Download: 243 GB
Context: 128K

Dense modelNeeds about 292 GB RAM

Meta's general models with tool calling and 128K context.

Tools

$ollama run llama3.1:405b

Qwen3 Coder 480B

Alibaba · Apache 2.0

Parameters: 480B
Download: 290 GB
Context: 256K

35B active per tokenNeeds about 348 GB RAM

Agentic coding models for long, multi-step tasks.

ToolsCode

$ollama run qwen3-coder:480b

DeepSeek-R1 671B

DeepSeek · MIT

Parameters: 671B
Download: 404 GB
Context: 160K

37B active per tokenNeeds about 485 GB RAM

Open reasoning models, from small distills to the full 671B.

ToolsThinking

$ollama run deepseek-r1:671b

DeepSeek-V3 671B

DeepSeek · DeepSeek License

Parameters: 671B
Download: 404 GB
Context: 160K

37B active per tokenNeeds about 485 GB RAM

A strong general mixture-of-experts model.

$ollama run deepseek-v3:671b