Local LLM local-llm.net
  • Learn
  • Articles
  • Directory
  • Community

Local LLM Models

62 models cataloged with VRAM requirements, setup commands, and use case recommendations. Browse by family or search for a specific model.

Llama

Llama 3.1

Meta's flagship open-weights LLM. Strong all-rounder with native tool use, 128K context, and broad ecosystem support.

8B chatinstructiontools

Llama 3.2

Meta's smallest Llama 3 models. 1B fits in 1GB VRAM; 3B is a sweet spot for low-end desktops and laptops.

1B chatsmallmobile

Llama 3.2 Vision

Multimodal Llama 3.2 with image input. 11B fits on consumer hardware; 90B needs 64GB+.

11B visionmultimodalimage

Llama 3.3

Meta's 70B offering with quality similar to Llama 3.1 405B at a fraction of the size.

70B chatinstructiontools

Llama 4

Meta's MoE multimodal Llama 4 generation. Scout has 10M context. Maverick matches GPT-4o on vision.

Scout 17B (109B MoE) visionmultimodalMoE

Code Llama

Meta's code-specialized Llama. Foundation for many code tools. Now superseded by Qwen 2.5-Coder and Qwen 3-Coder for new work.

7B codecompletionfill-in-middle

Llama 3.2 Vision

Meta's multimodal Llama 3.2. 11B is the consumer-friendly size; 90B is a flagship-tier VLM.

11B visionmultimodalimage

TinyLlama

1.1B Llama trained on 3T tokens. The smallest capable chat model. Runs on phones and Raspberry Pi.

1.1B chatsmalledge

Qwen

Qwen 2.5

Alibaba's flagship open weights. The strongest model in its size class for English, Chinese, code, math, and tool use.

0.5B chatinstructioncode

Qwen 2.5-Coder

Code-specialized Qwen 2.5. Strongest open-weight code model in its size class, supports fill-in-middle and chat.

0.5B codecompletionfill-in-middle

Qwen 3

Alibaba's Qwen 3 generation. Strongest all-round open model in 2025-2026. Includes 'thinking' mode for reasoning.

0.6B chatthinkingMoE

Qwen 3-Coder

Qwen 3 generation specialized for coding agents. 30B is competitive with 70B+ coders; 480B is a frontier-tier model.

30B codecode-agentslong-context

Qwen 3 VL

The most powerful vision-language model in the Qwen3 family. Combines Qwen 3's reasoning with state-of-the-art visual understanding. 2B-235B sizes for any hardware.

2B visionmultimodaltools

DeepSeek

DeepSeek R1

DeepSeek's reasoning model. Distilled 32B is excellent for 24GB GPUs. Full 671B MoE is frontier-class.

1.5B reasoningthinkingmath

DeepSeek V3

DeepSeek's flagship MoE. 671B total parameters but only 37B active. Needs 256GB+ for inference.

671B (37B active MoE) MoEgeneralcode

DeepSeek Coder

DeepSeek's code completion model. Strong for the size, especially at 6.7B and 33B.

1.3B codecompletion

DeepSeek Coder V2

DeepSeek Coder V2 open-source MoE code model. 16B active parameters with 236B total. Comparable to GPT-4-Turbo on code-specific tasks.

16B codeMoEcompletion

Mistral

Mistral

Mistral's first open model. Still useful for lightweight chat and edge deployment.

7B chatinstructionlightweight

Mistral Nemo

Mistral + NVIDIA collaboration. Excellent 12B with 128K context and tool use.

12B chatinstructiontools

Mistral Small

Mistral's sub-30B model. Best in class for its size; competitive with 70B+ on chat.

22B chattoolsinstruction

Mistral Large

Mistral's flagship. Frontier-class quality with 128K context. Needs data-center GPU or 2-3x consumer GPU.

123B chattoolsmultilingual

Gemma

Gemma 2

Google's open Gemma 2 family. Strong chat for the size; 2B is competitive with much larger models on simple tasks.

2B chatinstructionlightweight

Gemma 3

Gemma 3 with vision support. 12B is the sweet spot for single-GPU multimodal chat.

270M chatvisionmultilingual

Gemma 4

Google's frontier-tier Gemma 4 generation. Vision + tools + thinking + audio. Best single-GPU multimodal model in 2026.

e2b visiontoolsthinking

Phi

Phi-3

Microsoft's small Phi-3 family. Mini (3.8B) is the best-in-class small chat; Medium (14B) competes with much larger models.

Mini 3.8B chatlightweightsmall

Phi-4

Microsoft's Phi-4 generation. 14B with strong reasoning despite small size.

14B chatreasoningsmall

Phi-4 Mini

Phi-4-mini brings multilingual support, reasoning, and mathematics to a 3.8B model. Includes function calling. Strong on phones and edge devices.

3.8B chattoolsfunction-calling

Phi-4 Reasoning

Phi-4 reasoning and reasoning plus are 14B parameter open-weight reasoning models that rival much larger models on complex reasoning tasks.

14B reasoningmathsmall

StarCoder

StarCoder2

BigCode's StarCoder2 generation. 3B/7B/15B for code completion. Supports fill-in-middle and 600+ programming languages.

3B codecompletionfill-in-middle

Nomic

Nomic Embed Text

Nomic's open-weights embedding model. The default for many RAG pipelines. 137M parameters, runs on CPU.

137M embeddingretrievalRAG

MxBai

MxBai Embed Large

Mixedbread's open-weights embedding model. State-of-the-art quality at 335M parameters.

335M embeddingretrievalRAG

BGE

BGE-M3

BAAI's multilingual embedding model. Supports dense, sparse, and multi-vector retrieval. Best for multilingual RAG.

567M embeddingmultilingualRAG

LLaVA

LLaVA

LLaVA (Large Language and Vision Assistant). Open multimodal that combines a vision encoder with Vicuna/Llama/Mistral.

7B visionmultimodalimage

Moondream

Moondream

Tiny 1.8B vision-language model. Runs on phones and edge devices. Best small VLM in 2026.

1.8B visionsmalledge

SmolLM

SmolLM2

HuggingFace's small SmolLM2. 135M, 360M, and 1.7B sizes for edge and IoT. Trained on high-quality data.

135M chatsmalledge

OLMo

OLMo 2

Allen AI's fully open OLMo 2. Training data, code, and weights all released. Best for research reproducibility.

7B open-sourcechatresearch

Granite

Granite 3.3

IBM's enterprise-grade Granite. Apache 2.0, 128K context, strong on RAG and tool use. Designed for IBM customers but free to use.

2B enterprisecodetools

Command

Command R

Cohere's Command R. Strong RAG and tool use; 10+ language support. Best when you need grounded generation.

35B RAGtoolsmultilingual

Hermes

Hermes 3

Nous Research's Hermes 3. Strong function calling, supports system prompts, uncensored fine-tunes. Open weights, Apache 2.0.

3B chattoolsfunction-calling

Falcon

Falcon 3

TII's Falcon 3 family. 1B-10B sizes with strong quality-per-FLOP. Multilingual support.

1B chatlightweightmultilingual

GLM

GLM-5

Z.ai's flagship reasoning model. 744B total parameters, 40B active. Built for complex systems engineering and long-horizon agentic tasks. Frontier-class on SWE-Bench Pro.

744B (40B active MoE) reasoningthinkingagentic

GLM-5.1

GLM-5.1 is the next-generation flagship from Z.ai with significantly stronger coding capabilities. State-of-the-art on SWE-Bench Pro. Leads GLM-5 by a wide margin.

744B (40B active MoE) reasoningthinkingagentic

GLM-5.2

Latest Z.ai frontier model. Updated GLM-5.1 with improved agentic workflows and reasoning. One of the most-pulled models in the Ollama library in mid-2026.

744B (40B active MoE) reasoningthinkingagentic

GLM-4.7

Z.ai's GLM-4.7 generation. Advances coding capability over GLM-4.6 with stronger agentic tool use.

mid-size MoE codingagenticMoE

GLM-4.7 Flash

The strongest model in the 30B class. Balances performance and efficiency for lightweight local deployment. New option for the 24GB-VRAM sweet spot.

30B codinglightweighttools

Kimi

Kimi K2

Moonshot AI's Kimi K2. 1 trillion total parameters, only 32B active per token. Frontier-class agentic capability with strong tool use. One of the most-pulled models in the Ollama library in 2026.

1T (32B active MoE) MoEtoolsagentic

Kimi K2.5

Moonshot AI's January 2026 update to Kimi K2. Continued improvements in agentic capability and long-context understanding.

MoE MoEtoolsagentic

Kimi K2.6

Moonshot AI's March 2026 frontier update. Continued gains in agentic benchmarks and tool use. Among the most-pulled models in 2026.

MoE MoEtoolsagentic

Kimi K2.7 Code

Kimi K2.7 code-specialized variant. Strong on long-horizon coding agent workflows and multi-file refactors.

MoE codeMoEagentic

minimax

minimax M2

minimax's M2 series. Exceptional multilingual capabilities to elevate code engineering. Designed for real-world productivity and coding tasks across 100+ languages.

230B chattoolsmultilingual

minimax M2.1

minimax's December 2025 update to M2. Improved multilingual code engineering capabilities.

230B chattoolsmultilingual

minimax M2.5

minimax M2.5: state-of-the-art large language model designed for real-world productivity and coding tasks.

230B chattoolscoding

minimax M2.7

minimax M2-series model for coding, agentic workflows, and professional productivity. Among the most-pulled models in the Ollama library in 2026.

480B chattoolscoding

minimax M3

minimax M3, the latest M-series model. Continued improvements for coding, agentic workflows, and professional productivity.

230B chattoolscoding

Nemotron

Nemotron 3 Super

NVIDIA Nemotron 3 Super. 120B open MoE activating just 12B parameters. Delivers maximum compute efficiency and accuracy for complex multi-agent applications.

120B (12B active MoE) reasoningthinkingMoE

GPT-OSS

GPT-OSS

OpenAI's first open-weight model since GPT-2. Designed for powerful reasoning, agentic tasks, and versatile developer use cases. Strong function calling and code generation.

20B reasoningthinkingtools

Ministral

Ministral 3

Ministral 3 family designed for edge deployment. Capable of running on a wide range of hardware from Raspberry Pi to multi-GPU servers. Vision support in cloud variant.

3B chatedgevision

LFM2

LFM2.5 Thinking

LFM2.5 hybrid model with thinking capabilities at 1.2B. Designed for on-device deployment with reasoning. Tiny footprint, big thinking.

1.2B reasoningthinkingedge

QwQ

QwQ

QwQ is the reasoning model of the Qwen series. 32B size with thinking capability. Strong on math and logic problems.

32B reasoningthinkingmath

Magistral

Magistral

Magistral is a small, efficient reasoning model with 24B parameters. From Mistral AI. Best for hardware that cannot run larger reasoning models.

24B reasoningthinkingsmall

Devstral

Devstral

Devstral: the best open source model for coding agents. From Mistral AI. 24B size designed for software engineering tasks and multi-file agent workflows.

24B codetoolsagentic

Cogito

Cogito

Cogito v1 Preview is a family of hybrid reasoning models by Deep Cogito that outperform the best available open models of the same size across most standard benchmarks.

3B reasoninghybridtools
■ local-llm.net

The definitive guide to deploying AI locally. Run models on your own hardware with complete privacy and control.

A Srushta Media Limited project.

Learn

  • What Is Local AI?
  • Why Run AI Locally
  • Hardware Guide
  • Glossary

Tools

  • Tool Directory
  • Ollama
  • Mullama
  • Llamafu

Community

  • Blog
  • Community
  • GitHub
  • RSS

Company

  • About
  • Press Kit
  • Developers
  • Srushta Media

© 2026 Srushta Media Limited. All rights reserved.

RSS